1
00:00:04,001 --> 00:00:07,000
Hi, everyone.
My name is Chris Freeland and I'm the

2
00:00:07,000 --> 00:00:08,001
librarian at the Internet Archive.

3
00:00:09,001 --> 00:00:13,000
So here we are today at our final session
of the Library's Laboratory series.

4
00:00:14,000 --> 00:00:17,001
Over the past 10 weeks, 10 weeks if you
can believe it, we've brought together

5
00:00:17,001 --> 00:00:21,001
some of the world's leading scholars in
the digital humanities and data intensive

6
00:00:21,001 --> 00:00:25,000
sciences to talk about their projects
and how they're using the Internet

7
00:00:25,000 --> 00:00:26,001
Archive in the course of their research.

8
00:00:27,000 --> 00:00:31,000
We started off our series by asking the
bold question, what can you do with

9
00:00:31,000 --> 00:00:36,000
billions of archived web pages? And two
weeks ago, we ran out the formal part of

10
00:00:36,000 --> 00:00:40,000
our series by asking, what can you do
with 60 million digitized pages of

11
00:00:40,000 --> 00:00:45,000
scientific literature? In between,
we've heard from bibliographers, authors,

12
00:00:45,001 --> 00:00:49,001
educators, archivists, and data scientists
about how they're using materials,

13
00:00:50,000 --> 00:00:54,000
services, and infrastructure from the
Internet Archive. The recordings for all of

14
00:00:54,000 --> 00:00:58,000
those sessions are now available online.
And I see that Duncan, who's working

15
00:00:58,000 --> 00:01:02,000
behind the scenes today, along with Caitlin,
hello both of you and thank you, has

16
00:01:02,000 --> 00:01:03,001
shared that out into the chat.

17
00:01:05,000 --> 00:01:08,000
But we're going to do something a
little different today. When we started

18
00:01:08,000 --> 00:01:12,001
planning this series, we had inquiries
from a number of scientists who wanted to

19
00:01:12,001 --> 00:01:16,001
do shorter presentations, you know,
give something more like a status update or

20
00:01:16,001 --> 00:01:22,001
an overview of the research, not a 45-minute
lecture with a long Q&A. So we put

21
00:01:22,001 --> 00:01:27,000
out a call for lightning talks.
And today we're going to bring you a variety of

22
00:01:27,000 --> 00:01:32,000
short talks and videos from researchers
working on topics as varied as

23
00:01:32,000 --> 00:01:36,000
understanding the effects of emotional
cues by the chairs of the U.S. Federal

24
00:01:36,000 --> 00:01:39,001
Reserve on financial markets
to a DIY book scanning

25
00:01:39,001 --> 00:01:42,001
robot. So here's the game plan for today.

26
00:01:43,001 --> 00:01:47,001
Live transcripts are available, captions
are available. Use the live transcripts

27
00:01:47,001 --> 00:01:54,001
feature of Zoom to turn those on.
We also, you can copy from within the chat

28
00:01:54,001 --> 00:01:58,000
by just mousing over anything
individual that you're

29
00:01:58,000 --> 00:02:00,000
interested,
like links that we're going to share.

30
00:02:00,000 --> 00:02:04,000
And also, we're going to capture all of
the links in the chat and make those

31
00:02:04,000 --> 00:02:09,001
available, including the video, all the
resources will be available. So if you're

32
00:02:09,001 --> 00:02:14,000
trying to catch something, if you want
to remember something, copy a link that

33
00:02:14,000 --> 00:02:18,000
we've shared, rest assured, that will make
it into the email that we're going to

34
00:02:18,000 --> 00:02:23,000
send you tomorrow with the video for today's
session and the other resources that

35
00:02:23,000 --> 00:02:29,000
we've shared. As you see the chat is open,
please do be respectful and keep the

36
00:02:29,000 --> 00:02:35,000
comments on topic and use the Q&A feature
to submit questions for our panelists.

37
00:02:35,000 --> 00:02:42,000
And we'll have time for probably one
question per speaker. So do please submit

38
00:02:42,000 --> 00:02:46,000
questions for us to gather from. And a final

39
00:02:46,000 --> 00:02:47,001
thing I do want to mention about time.

40
00:02:48,000 --> 00:02:52,000
I anticipate that we're definitely going
to run over an hour today, probably more

41
00:02:52,000 --> 00:02:57,001
towards 90 minutes. So for those of you
who need to depart at the top of the

42
00:02:57,001 --> 00:03:02,001
hour, rest assured, we're going to
record all of this and it will be made

43
00:03:02,001 --> 00:03:08,000
available to everyone. So for now,
I can see people are already using the chat.

44
00:03:08,000 --> 00:03:12,000
Please do say hello. Let us know who you
are and where you're joining in from

45
00:03:12,000 --> 00:03:18,001
today. So I actually was just looking back
over my planning notes and we started

46
00:03:18,001 --> 00:03:24,000
talking about what became this Library
as Laboratory series last September. And

47
00:03:24,000 --> 00:03:29,000
like many ideas at the Internet Archive,
it burst forth with enthusiasm and gusto

48
00:03:29,000 --> 00:03:33,000
in a meeting with Brewster Cale,
the founder and digital librarian of the

49
00:03:33,000 --> 00:03:37,000
Internet Archive. So I'd like to welcome
Brewster to the screen. And Brewster,

50
00:03:37,001 --> 00:03:41,001
I'd like to have you share a bit of
your thinking. Why was organizing this

51
00:03:41,001 --> 00:03:46,000
Digital Humanities Expo a priority
for you and for the Internet

52
00:03:46,000 --> 00:03:49,000
Archive? Thank you, Chris.

53
00:03:54,000 --> 00:03:58,001
This is really fulfilling the dream of the
Internet Archive. We started by trying

54
00:03:58,001 --> 00:04:02,001
to archive the Internet and how do you
go and do that and then expand it to

55
00:04:02,001 --> 00:04:06,000
archiving other things. But it wasn't just
to make it so that you can go and find

56
00:04:06,000 --> 00:04:09,001
old web pages.
It was to try to get a bigger view.

57
00:04:10,000 --> 00:04:15,001
Can we go and help make it so that lots
of new and different things can happen

58
00:04:15,001 --> 00:04:20,001
without having to build your own collection
yourself, build a library that you

59
00:04:20,001 --> 00:04:25,000
can just go and have your new idea and
all of the materials needed for your

60
00:04:25,000 --> 00:04:30,000
research are on the shelves, but now just
digital shelves. So that was sort of

61
00:04:30,000 --> 00:04:34,001
the impetus for the Internet Archive
in general. And so this idea of having

62
00:04:34,001 --> 00:04:38,001
people be able to use these collections
at scale has been so important, but

63
00:04:38,001 --> 00:04:43,001
actually it's really pretty hard.
The collections are, I think, in pretty good

64
00:04:43,001 --> 00:04:49,000
shape, but they're just huge and hard to
find and figuring out how to go and use

65
00:04:49,000 --> 00:04:55,001
them. But we have seen fabulous presenters
and including the ones that did

66
00:04:55,001 --> 00:05:00,001
today. But the urgency of misinformation
and feeling that our information

67
00:05:00,001 --> 00:05:06,000
ecosystem is out of our control.
I think we need, we really need, people with a

68
00:05:06,000 --> 00:05:11,000
macroscope. This is Jesse Ossabell's line.
He said, we got really far with a

69
00:05:11,000 --> 00:05:16,000
microscope towards understanding and humans
and scientists and science got really

70
00:05:16,000 --> 00:05:23,000
further along. Now we need a bigger
picture of what's going on. Can we use web

71
00:05:23,000 --> 00:05:28,001
scale, TV scale, book scale types
information to try to help us understand our

72
00:05:28,001 --> 00:05:32,000
world. And I'm jazzed that many of the
key players have come forward to speak

73
00:05:32,000 --> 00:05:37,000
about what they're doing. They're so
inspiring and inspiring to me and then

74
00:05:37,000 --> 00:05:41,000
they're inspiring to our staff,
but I think others to go and say, yes, I can go

75
00:05:41,000 --> 00:05:46,001
and use these sorts of collections at
scale to go and do different things. So

76
00:05:46,001 --> 00:05:51,000
what's the value to the Internet Archive
of this? Just from a completely selfish

77
00:05:51,000 --> 00:05:55,000
point of view, it helps drive us forward
to make more useful collections and

78
00:05:55,000 --> 00:06:01,001
tools. And so for that, we're going to
need your feedback as to what the Internet

79
00:06:01,001 --> 00:06:06,001
Archive can do to be a better library.
What are the materials, the tools, the

80
00:06:06,001 --> 00:06:12,000
structures that you would like to see or
platforms? Is this library as laboratory

81
00:06:12,000 --> 00:06:18,001
series useful to you? We've seen a large
number of people join in on these.

82
00:06:19,000 --> 00:06:26,000
So, but please, feedback. That's absolutely
essential. And thank you, Chris, for

83
00:06:26,000 --> 00:06:31,001
making all of this come about with a
tremendous group of researchers that have

84
00:06:31,001 --> 00:06:37,001
come together. So looking forward to today.
Thanks for that, Brewster. Always

85
00:06:37,001 --> 00:06:43,001
good to hear from you and hear what you're
thinking about. So as we jump in here

86
00:06:43,001 --> 00:06:48,001
today, I want to let all of you in the
audience know what to expect. So we have

87
00:06:48,001 --> 00:06:53,000
three segments of talks. And each segment
will have three to four talks. That's

88
00:06:53,000 --> 00:06:58,000
talks and videos. So what we'll do is
we'll move from talk to talk quickly, and

89
00:06:58,000 --> 00:07:03,000
we'll wrap up each segment with a quick
round of questions and answers. So if you

90
00:07:03,000 --> 00:07:07,001
have questions, again, drop them into the
Q&A so that we can we'll probably, as I

91
00:07:07,001 --> 00:07:12,001
said, have time to ask one question aloud.
But our speakers are going to be

92
00:07:12,001 --> 00:07:17,000
hanging around. And so if you drop off
additional questions, they might be

93
00:07:17,000 --> 00:07:21,000
interested in engaging with you further
and answering some of those questions

94
00:07:21,000 --> 00:07:27,001
from the from the Q&A. So please,
as you have questions, please do ask them using

95
00:07:27,001 --> 00:07:34,001
the Q&A feature. I'll also mention that
Duncan dropped a link into the chat

96
00:07:34,001 --> 00:07:39,001
with the agenda so that you can follow
along the order that we're going in today.

97
00:07:40,000 --> 00:07:45,001
So let's get started. Up first today
is Kate Miltner from the University of

98
00:07:45,001 --> 00:07:49,001
Edinburgh. Kate's going to tell us about
the forgotten histories of the mid

99
00:07:49,001 --> 00:07:52,001
-century coding boot camp. Over to you, Kate.

100
00:07:55,001 --> 00:07:59,001
Hi, everyone. Thanks so much for joining.
Just going to share my screen and we

101
00:07:59,001 --> 00:08:06,001
will get this party started. Okay, so

102
00:08:06,001 --> 00:08:11,001
as Chris mentioned, I'm Kate Miltner.
I am a Marie Curie postdoctoral fellow at

103
00:08:11,001 --> 00:08:15,000
the University of Edinburgh.
And I'm just really grateful to Chris and the

104
00:08:15,000 --> 00:08:19,000
Internet Archive for having me to talk
about some of the amazing artifacts that

105
00:08:19,000 --> 00:08:24,000
I've used in my historical research on
electronic data processing schools, which

106
00:08:24,000 --> 00:08:29,000
are what I consider to be the mid-century
predecessor of the contemporary coding

107
00:08:29,000 --> 00:08:35,000
boot camp. So in the past decade,
news stories like these have become

108
00:08:35,000 --> 00:08:40,000
increasingly familiar. I'm sure most,
if not all of you, have seen or read an

109
00:08:40,000 --> 00:08:43,000
article that talks about
learning to code and what it can

110
00:08:43,000 --> 00:08:45,001
accomplish. Maybe you even wrote one of them.

111
00:08:46,001 --> 00:08:51,001
As a part of my PhD thesis, I read over
200 articles over a 10-year period that

112
00:08:51,001 --> 00:08:56,000
talked about learning to code. In reading
all of these articles, it became clear

113
00:08:56,000 --> 00:09:00,000
that coding has been positioned as a
solution to a variety of interconnected

114
00:09:00,000 --> 00:09:05,001
social issues, including concerns about AI,
gender and racial bias in the tech

115
00:09:05,001 --> 00:09:11,000
sector, economic inequality, and skills
gaps that have supposedly left millions

116
00:09:11,000 --> 00:09:15,000
of well-paid positions unfilled due
to a lack of appropriate training.

117
00:09:16,001 --> 00:09:21,000
In response to this discourse,
an entire industry of coding boot camps has

118
00:09:21,000 --> 00:09:26,000
developed in the US and across the world.
Coding boot camps are short-term

119
00:09:26,000 --> 00:09:30,000
intensive courses that aim to make
professional technologists out of technical

120
00:09:30,000 --> 00:09:36,000
novices. The topics of these programs can vary.
Some are focused on data science,

121
00:09:36,001 --> 00:09:41,000
some on user experience, some on software
engineering. There's even programs for

122
00:09:41,000 --> 00:09:45,000
digital marketing and product management.
The length of time can vary too. Most

123
00:09:45,000 --> 00:09:49,001
are around three to four months,
but some even last as long as two years. But

124
00:09:49,001 --> 00:09:53,001
independent of the topic and the timeline,
the promises that these programs make

125
00:09:53,001 --> 00:09:58,000
are usually the same. Stick with us and
the tech career of your dreams will be

126
00:09:58,000 --> 00:10:04,000
right within reach. It's a pretty compelling
promise. Across the US and Canada

127
00:10:04,000 --> 00:10:09,001
alone, the coding boot camp industry trained
almost 25,000 people a year in 2020,

128
00:10:10,000 --> 00:10:16,000
pulling in almost 350 million dollars.
One of the claims made by coding boot

129
00:10:16,000 --> 00:10:20,000
camps is that they offer a novel solution
for addressing some of the major

130
00:10:20,000 --> 00:10:24,001
problems within tech education.
First, they're a lot shorter than a four-year

131
00:10:24,001 --> 00:10:28,001
college degree. Second, they're supposed
to have a much lower sticker price.

132
00:10:29,000 --> 00:10:32,001
Third, they're supposed to be more
accessible to groups that are traditionally

133
00:10:32,001 --> 00:10:37,000
excluded from the tech industry.
And finally, they're more agile than a

134
00:10:37,000 --> 00:10:40,001
university, which is supposed to make
them responsive to the needs of industry.

135
00:10:41,000 --> 00:10:45,001
With all of these factors combined,
they're meant to be a new and ideal way to

136
00:10:45,001 --> 00:10:51,001
respond to the challenges of what some
call the fourth industrial revolution. But

137
00:10:51,001 --> 00:10:55,001
as any tech scholar will tell you,
the likelihood of something tech related from

138
00:10:55,001 --> 00:11:00,001
the current moment being completely brand
new is pretty unlikely. So I looked to

139
00:11:00,001 --> 00:11:05,000
find the historical roots of the coding
boot camp. And what I was able to find

140
00:11:05,000 --> 00:11:07,000
was surprising, even to me.

141
00:11:08,001 --> 00:11:12,000
As I was reading through some work in
the history of computing, I came across

142
00:11:12,000 --> 00:11:17,000
Nathan Entsmanger's fantastic book about
the software industry in the mid 20th

143
00:11:17,000 --> 00:11:21,001
centuries. In it, he wrote about the
prevalence of data, sorry, electronic data

144
00:11:21,001 --> 00:11:28,000
processing schools or EDP schools in
the 1960s and 70s. EDP schools were

145
00:11:28,000 --> 00:11:31,001
vocational schools that offered short-term
courses in computer programming.

146
00:11:32,000 --> 00:11:36,000
And much like coding boot camps,
these privately run schools aim to train people

147
00:11:36,000 --> 00:11:42,000
for jobs in industry. After reading about
EDP schools, I started looking for some

148
00:11:42,000 --> 00:11:47,000
primary historical materials about them.
Based on the sources available to me at

149
00:11:47,000 --> 00:11:51,001
the university where I did my PhD,
I was able to find a few newspaper articles. I

150
00:11:51,001 --> 00:11:55,000
also found ads for one franchise of
schools in particular, which was the

151
00:11:55,000 --> 00:11:58,001
Electronic Computer Programming
Institute or the ECPI.

152
00:11:59,000 --> 00:12:03,001
In the ads especially, I began to see some
similarities between the sales pitches

153
00:12:03,001 --> 00:12:08,000
of EDP schools and the sales pitches of
coding boot camps, especially around how

154
00:12:08,000 --> 00:12:12,001
easy programming can be and the number
of available jobs in the computing

155
00:12:12,001 --> 00:12:18,000
industry. I was hoping to find some
more materials about EDP schools in

156
00:12:18,000 --> 00:12:21,001
traditional archives like the Charles
Babbage Institute at the University of

157
00:12:21,001 --> 00:12:26,000
Minnesota. But because most EDP schools
were either regional, short-lived, or

158
00:12:26,000 --> 00:12:30,000
both, the finding aids for the archives
suggested that there wasn't a lot of

159
00:12:30,000 --> 00:12:31,001
archival material to be found.

160
00:12:32,001 --> 00:12:37,001
But then, thanks to the junk mail
collection of a computing pioneer named Ted

161
00:12:37,001 --> 00:12:42,000
Nelson and the wonders of digitization,
I stumbled upon something that I thought

162
00:12:42,000 --> 00:12:47,001
I'd never find, which was an original
recruitment brochure from the ECPI from the

163
00:12:47,001 --> 00:12:54,000
late 1960s. The ECPI booklet was only 16
pages long, but it was a crucial piece

164
00:12:54,000 --> 00:12:58,000
of evidence in my historical research
that allowed me to make important links

165
00:12:58,000 --> 00:13:04,001
between the past and present of computing
education. It was really remarkable how

166
00:13:04,001 --> 00:13:09,000
many similarities there were between how
EDP schools and coding boot camps framed

167
00:13:09,000 --> 00:13:15,000
themselves and what they have to offer.
First, both kinds of organizations framed

168
00:13:15,000 --> 00:13:19,000
a career in computing as accessible
to anyone and a good way to guard

169
00:13:19,000 --> 00:13:20,001
against the threat of automation.

170
00:13:22,001 --> 00:13:27,001
EDP schools also focused on the wide
availability of computing jobs, much like

171
00:13:27,001 --> 00:13:33,001
contemporary discussions of the skills gap.
They also focused on computing as an

172
00:13:33,001 --> 00:13:38,000
inclusive career pathway for women and
people of color. This was actually pretty

173
00:13:38,000 --> 00:13:42,000
remarkable for the 1960s.
Out of the 10 students showcased for successful

174
00:13:42,000 --> 00:13:46,001
placements, over half of them were
women or people from minoritized ethnic

175
00:13:46,001 --> 00:13:52,001
groups. Finally, the ECPI booklet
underscored their links with industry, much

176
00:13:52,001 --> 00:13:58,000
like coding boot camps do today.
Of course, the image that an organization

177
00:13:58,000 --> 00:14:03,001
presents to the world and its reality can
be starkly different. So I then began

178
00:14:03,001 --> 00:14:07,001
to look into whether the image that ECPI
presented connected with other parts of

179
00:14:07,001 --> 00:14:13,000
the historical record. To do this,
I turned to another fantastic Internet Archive

180
00:14:13,000 --> 00:14:17,001
resource, which was their digitized
collection of computer world. To have this as

181
00:14:17,001 --> 00:14:21,001
a searchable digital archive was really
incredible because it contained some

182
00:14:21,001 --> 00:14:27,000
remarkably rich material that I probably
wouldn't have found otherwise. One of

183
00:14:27,000 --> 00:14:31,000
the key directions that the computer world
archive encouraged me to pursue in my

184
00:14:31,000 --> 00:14:35,001
research was the disconnect between the
claims of inclusivity made about the

185
00:14:35,001 --> 00:14:39,001
computing industry at the time and the
reality of the computing labor market.

186
00:14:40,001 --> 00:14:44,001
There was a ton of coverage in computer
world about EDP schools in the 1960s and

187
00:14:44,001 --> 00:14:50,000
70s, and a lot of it highlighted many of
the issues with these organizations. A

188
00:14:50,000 --> 00:14:54,001
series of articles published from 1969
to 1970 showed how these claims of

189
00:14:54,001 --> 00:14:59,001
accessibility were not necessarily so true.
Despite a purported programmer

190
00:14:59,001 --> 00:15:03,001
shortage, these articles illustrated how
Black programming trainees found it next

191
00:15:03,001 --> 00:15:08,000
to impossible to get hired for programming
jobs. This was a very different story

192
00:15:08,000 --> 00:15:10,000
than the one told by the ECPI booklet.

193
00:15:11,001 --> 00:15:15,001
What the computer world archive also
highlighted was how problematic

194
00:15:15,001 --> 00:15:17,000
so many of these schools were.

195
00:15:17,001 --> 00:15:22,000
In Nathan Ensminger's book, he discussed
how many companies had employed a no-EDP

196
00:15:22,000 --> 00:15:26,001
school graduate policy due to the variable
quality of these schools. Articles

197
00:15:26,001 --> 00:15:30,001
that I found in computer world gave clear
insight into how these organizations

198
00:15:30,001 --> 00:15:37,001
operated and the dire situations that
some of their students were left in. So

199
00:15:37,001 --> 00:15:41,001
you may be asking yourself, okay,
so what does the history of EDP schools in the

200
00:15:41,001 --> 00:15:46,001
1960s and 70s have to do with today?
There's a reason that this history is worth

201
00:15:46,001 --> 00:15:49,001
paying attention to,
and that's because it threatens to repeat

202
00:15:49,001 --> 00:15:51,001
itself in the current moment.

203
00:15:53,000 --> 00:15:57,000
The ECPI booklet showed us that there
are a lot of similarities between how

204
00:15:57,000 --> 00:16:01,000
coding bootcamps and EDP schools
presented both themselves and the

205
00:16:01,000 --> 00:16:02,001
potential benefits of learning to code.

206
00:16:03,001 --> 00:16:06,000
But the computer world archive
shows us that there may be

207
00:16:06,000 --> 00:16:07,001
some other similarities as well.

208
00:16:08,001 --> 00:16:13,000
The story of EDP schools didn't end well,
either for many of the students or for

209
00:16:13,000 --> 00:16:18,000
the schools themselves. They were shut
down and liquidated by the mid-1970s after

210
00:16:18,000 --> 00:16:22,000
a federal investigation cracked down on
fraudulent vocational schools. Others

211
00:16:22,000 --> 00:16:26,001
just went out of business because no one
wanted to go anymore. Of course, the

212
00:16:26,001 --> 00:16:30,000
future of coding bootcamps has yet to
be written, but there are some warning

213
00:16:30,000 --> 00:16:34,000
signs that there might be more similarities
between bootcamps and EDP schools

214
00:16:34,000 --> 00:16:38,000
than we might hope. One of my goals for
this project was to point out how the

215
00:16:38,000 --> 00:16:42,000
past lives on in the present so that the
mistakes of the past can hopefully be

216
00:16:42,000 --> 00:16:47,001
avoided. If you've enjoyed this talk,
I have a journal article coming out this

217
00:16:47,001 --> 00:16:51,001
fall in Information at Culture that talks
about the commonalities between coding

218
00:16:51,001 --> 00:16:55,001
bootcamps and EDP schools at length.
So please do keep an eye out.

219
00:16:57,000 --> 00:17:00,000
Thanks so much for your attention.
And if you'd like to get in touch or put

220
00:17:00,000 --> 00:17:07,000
questions in the Q&A, please do.
Thanks so much, Kate. Really appreciate that

221
00:17:07,000 --> 00:17:11,000
excellent overview. We have some questions
that have come in. I'm gathering

222
00:17:11,000 --> 00:17:16,000
those. If others in the audience have
questions, please do use the Q&A feature to

223
00:17:16,000 --> 00:17:22,000
drop those off. And what we'll do now is
move on to the next talk. And so up next

224
00:17:22,000 --> 00:17:28,001
is Tom Galli from the University of Tokyo.
Now, Tom is in Japan, and he will be

225
00:17:28,001 --> 00:17:33,000
watching this time shifted.
So this is a wave to Tom in the future.

226
00:17:33,000 --> 00:17:34,001
He will be watching this in the past.

227
00:17:35,000 --> 00:17:41,000
I'm getting a little confused,
but let's let Tom explain in this video some of

228
00:17:41,000 --> 00:17:48,000
his research about Japan.
So, Caitlin, you want to take it away? Japan is

229
00:17:48,000 --> 00:17:54,000
now a modern country, not too different
from many other places. But when it

230
00:17:54,000 --> 00:17:59,001
opened up to the world in the middle of
the 19th century, Japan seemed exotic and

231
00:17:59,001 --> 00:18:04,001
mysterious to the first Western visitors.
People who couldn't visit were even

232
00:18:04,001 --> 00:18:10,001
more curious about the country.
So to satisfy that curiosity, and to make some

233
00:18:10,001 --> 00:18:15,001
money, those visitors wrote books
about Japan for the people back home.

234
00:18:17,000 --> 00:18:22,000
Japan, as they saw it, is a collection
of excerpts from those books. They show

235
00:18:22,000 --> 00:18:27,000
how those visitors, mostly American or British,
describe the country and its

236
00:18:27,000 --> 00:18:34,000
people. The cities, the countryside,
the clothing, the religions, the

237
00:18:34,000 --> 00:18:37,000
theater and festivals, even the smells.

238
00:18:38,001 --> 00:18:44,000
They tell about writing in rickshaws,
attending weddings, meeting prostitutes,

239
00:18:44,001 --> 00:18:51,000
experiencing earthquakes. I got the idea
to create Japan as they saw it, after I

240
00:18:51,000 --> 00:18:54,000
came across some of those old
books at the Internet Archive.

241
00:18:56,000 --> 00:19:00,001
You see, I myself moved to
Japan in 1983 from California,

242
00:19:01,001 --> 00:19:03,000
and I've lived here ever since.

243
00:19:04,000 --> 00:19:09,001
So it was fascinating for me to think
about how my own first impressions of the

244
00:19:09,001 --> 00:19:15,001
country, and what I had told my friends
and family back home about it, compared

245
00:19:15,001 --> 00:19:20,001
with what people had written about
Japan a century earlier. I thought a

246
00:19:20,001 --> 00:19:25,000
collection of those earlier impressions
might be interesting for others to read

247
00:19:25,000 --> 00:19:31,001
too. So I first prepared a list of
over 240 books about Japan at the

248
00:19:31,001 --> 00:19:36,000
Internet Archive,
published between 1855 and 1912.

249
00:19:37,001 --> 00:19:42,000
I went through those books and chose
passages that seemed interesting or amusing,

250
00:19:42,001 --> 00:19:47,000
and I put them all on this website.
It's also available as an e-book.

251
00:19:48,001 --> 00:19:53,000
Each excerpt is linked to the original
book at the Internet Archive. And like

252
00:19:53,000 --> 00:19:57,001
those books at the Internet Archive,
Japan as they saw it is free for

253
00:19:57,001 --> 00:19:59,000
anyone in the world to read.

254
00:20:00,001 --> 00:20:04,001
Thank you once again, Internet Archive,
for making this possible.

255
00:20:09,000 --> 00:20:14,000
And a thanks to you, Tom, for putting
that video together. And if anyone has a

256
00:20:14,000 --> 00:20:17,000
question, yes. We've already reached out
to Tom and asked him if we would create

257
00:20:17,000 --> 00:20:20,001
other videos for other parts of our collection,
because I thought that was really

258
00:20:20,001 --> 00:20:27,000
great. Also stay tuned. Tom is going to
close out our show with another video on

259
00:20:27,000 --> 00:20:31,000
forgotten novels of the 19th century,
which is really fun research. But up next,

260
00:20:31,000 --> 00:20:36,000
we have one of my favorite people on the
planet, a colleague who I've worked with

261
00:20:36,000 --> 00:20:41,000
for years in a variety of ways,
and I'm really pleased to welcome him to the

262
00:20:41,000 --> 00:20:47,001
stage virtually to tell us about his research.
So that's Rod Page from the

263
00:20:47,001 --> 00:20:51,001
University of Glasgow, talking about the
bibliography of life. Over to you, Rod.

264
00:21:01,001 --> 00:21:06,001
Okay. Thank you very much for that lovely
introduction, Chris. So I'm going to

265
00:21:06,001 --> 00:21:11,000
talk about the bibliography of life.
So I guess what I should do is

266
00:21:11,000 --> 00:21:13,000
define what I mean by that.

267
00:21:14,000 --> 00:21:19,000
So a number of people who work in the field
of biodiversity and taxonomy have the

268
00:21:19,000 --> 00:21:24,001
stream of the bibliography of life.
And basically, it's access to every taxonomic

269
00:21:24,001 --> 00:21:29,001
paper published on every species ever
described. So to give you a sense of the

270
00:21:29,001 --> 00:21:33,001
scale, we think there are probably about
two million species described on the

271
00:21:33,001 --> 00:21:37,000
planet today. There's probably about
10 million in total, so lots to be

272
00:21:37,000 --> 00:21:41,001
discovered. So we're thinking in terms
of hundreds of thousands, perhaps even a

273
00:21:41,001 --> 00:21:46,001
million or so publications that describe
these species. Now sometimes when we

274
00:21:46,001 --> 00:21:51,001
talk about this notion of a bibliography
of life, some people get upset because

275
00:21:51,001 --> 00:21:56,001
it seems to focus on taxonomy and what's
so special about taxonomy. What about

276
00:21:56,001 --> 00:22:01,000
all these major biomedical databases
such as PubMed, for example, lots of

277
00:22:01,000 --> 00:22:05,001
information on medicine and so on.
I want to make a case that there's something

278
00:22:05,001 --> 00:22:12,001
special about taxonomy. Taxonomy and
biodiversity in many ways, it's not

279
00:22:12,001 --> 00:22:15,000
big data, it's long data.

280
00:22:15,001 --> 00:22:19,001
We have lots and lots of long tales.
What you can see in the slide here on the

281
00:22:19,001 --> 00:22:25,001
right is a little summary of
the size of pages in Wikipedia

282
00:22:25,001 --> 00:22:30,001
for different mammal species.
So some mammals, for example, lions and mice,

283
00:22:31,000 --> 00:22:35,001
charismatic animals or medically important
animals, have really large Wikipedia

284
00:22:35,001 --> 00:22:40,001
entries. Down the bottom of this chart
you can see there are literally thousands

285
00:22:40,001 --> 00:22:44,000
of pages of Wikipedia on these
mammals that are very, very small.

286
00:22:45,000 --> 00:22:49,001
So for many species on the planet,
almost all that we know about them in terms of

287
00:22:49,001 --> 00:22:52,001
their ecology, their morphology,
what they do, where they are,

288
00:22:53,000 --> 00:22:54,001
will come from the taxonomic literature.

289
00:22:55,001 --> 00:23:02,001
Now, taxonomy itself is also very
much a subject of a long tale. So

290
00:23:02,001 --> 00:23:08,000
we have some very large, prominent journals
such as Zutaxa that publish tens of

291
00:23:08,000 --> 00:23:09,001
thousands of new species descriptions.

292
00:23:10,001 --> 00:23:15,001
But also there is a very, very long tale,
in many cases of often very small

293
00:23:15,001 --> 00:23:20,000
journals, that are often very
niche in terms of the taxonomic

294
00:23:20,000 --> 00:23:21,001
focus or the geographic focus.

295
00:23:21,001 --> 00:23:25,001
And again, they will be a reservoir of
lots of information about these species.

296
00:23:26,000 --> 00:23:29,001
So again, if we're going to get a nice
comprehensive bibliography of life, we're

297
00:23:29,001 --> 00:23:35,001
going to have to go hunting for those.
Now, those of you, I guess like myself and

298
00:23:35,001 --> 00:23:40,000
Chris have been around for a while in this
area, you might be thinking, hang on a

299
00:23:40,000 --> 00:23:43,001
second, this sounds already familiar.
What about the Biodiversity Heritage

300
00:23:43,001 --> 00:23:47,001
Library, the BHL? Isn't this what they're
doing? And we had a really interesting

301
00:23:47,001 --> 00:23:52,001
presentation a couple of weeks ago about them.
Well, there is some overlap, but

302
00:23:52,001 --> 00:23:57,000
BHL suffers from what I'm going to call
the Mickey Mouse Gap, which is this thing

303
00:23:57,000 --> 00:24:01,001
that the impact of copyright has had a
huge sort of dampening effect on BHL's

304
00:24:01,001 --> 00:24:06,001
coverage. What I'm trying to capture in
this diagram here is in blue, you can see

305
00:24:06,001 --> 00:24:10,000
for each decade, how many publications
are out there describing new species of

306
00:24:10,000 --> 00:24:15,000
animals, just a sort of lower balance on that.
And in red is the articles in BHL.

307
00:24:16,001 --> 00:24:21,001
And you can see that the coverage in BHL
dips dramatically after about 1923, when

308
00:24:21,001 --> 00:24:26,001
copyright in America kicks in,
and it sort of starts to increase a bit. But

309
00:24:26,001 --> 00:24:30,001
there's this big area in blue,
and that's the stuff that I'm after. All these

310
00:24:30,001 --> 00:24:36,001
descriptions of species, mostly in the
20th century, that aren't in BHL. So one

311
00:24:36,001 --> 00:24:42,000
solution is to try and capture these
articles and put them somewhere safe. Now, a

312
00:24:42,000 --> 00:24:46,000
lot of this material has actually been
digitized and is freely available in

313
00:24:46,000 --> 00:24:49,001
various websites. So I spent some time
trying to grab these together. And I've

314
00:24:49,001 --> 00:24:54,000
sort of created almost a mini BHL on
the Internet Archive collecting these

315
00:24:54,000 --> 00:24:57,001
publications. They're full of beautiful
photographs of species, also information

316
00:24:57,001 --> 00:25:02,001
on geography and maps and climate,
and also the people who study these species.

317
00:25:03,001 --> 00:25:06,001
So that's one approach, is just to try
and gather as many of these publications

318
00:25:06,001 --> 00:25:13,001
as possible. I briefly talked about the
long tail of taxonomic publications,

319
00:25:13,001 --> 00:25:17,001
or these small journals. Many of these
journals are as endangered as species or

320
00:25:17,001 --> 00:25:21,000
indeed taxonomists themselves. There are
journals that just simply vanish, like

321
00:25:21,000 --> 00:25:25,001
this one on the left, the recent Italian
journal that's just gone. There are also

322
00:25:25,001 --> 00:25:31,001
journals that vanish but sort of come back
as zombies, taken over by bad people.

323
00:25:32,001 --> 00:25:37,000
On the right, there's a journal that used
to publish on butterfly taxonomy. The

324
00:25:37,000 --> 00:25:41,000
person who ran that eventually sort of gave up,
it was too hard to sort of

325
00:25:41,000 --> 00:25:45,001
maintain a taxonomic journal. Somebody
else came along, took that domain name,

326
00:25:46,000 --> 00:25:50,001
and is now publishing things that anything
but butterfly taxonomy. And this is

327
00:25:50,001 --> 00:25:54,001
where the Wayback Machine has been
incredibly useful to try and retrieve these

328
00:25:54,001 --> 00:25:58,000
old journals and their content and also
discover the history of some of these

329
00:25:58,000 --> 00:26:03,001
journals which have been hijacked.
So part of this exercise is just getting the

330
00:26:03,001 --> 00:26:09,001
content. One thing that in a sense,
Internet Archive probably isn't particularly

331
00:26:09,001 --> 00:26:13,001
great at is metadata. You can put quite
a bit in, but for really detailed

332
00:26:13,001 --> 00:26:18,001
metadata, I've turned to Wikidata.
So Wikidata has this extraordinary resource.

333
00:26:18,001 --> 00:26:23,000
It's a great editing interface,
and there's an enormous community of people

334
00:26:23,000 --> 00:26:26,001
contributing. So for example,
if you have an article like this one, it's in

335
00:26:26,001 --> 00:26:31,000
English and Chinese, you can have titles
in both languages. But I think it really

336
00:26:31,000 --> 00:26:36,001
motivates my use of Wikidata is to try
and connect all these things together. So

337
00:26:36,001 --> 00:26:41,001
I guess what I'm aiming for ultimately
is something a bit like this. There's a

338
00:26:41,001 --> 00:26:46,000
link, I think this is also going to be
in the show notes. This is a little app

339
00:26:46,000 --> 00:26:51,001
that I built that essentially talks to
Wikidata, and you can look at information,

340
00:26:51,001 --> 00:26:56,001
say, on species on the left. You can get
information on journals, publish papers

341
00:26:56,001 --> 00:27:00,001
on these species. And you can also then
bounce to the taxonomist that people

342
00:27:00,001 --> 00:27:04,001
actually doing the research.
So the goal is to have this kind of network of

343
00:27:04,001 --> 00:27:09,000
linked things, images of species,
journal articles, and the taxonomist

344
00:27:09,000 --> 00:27:12,001
themselves. And these little thumbnails
you can see in these pictures here,

345
00:27:13,000 --> 00:27:16,001
they're all coming from Internet Archive.
So these are articles that are freely

346
00:27:16,001 --> 00:27:22,000
available that people can read and see
pictures of these species. So we're next

347
00:27:22,000 --> 00:27:28,000
for this project. I guess I'm fairly simple
minded in a way. One of it is just to

348
00:27:28,000 --> 00:27:31,001
try and get more content and get as much
of that content digitally preserved as

349
00:27:31,001 --> 00:27:36,000
possible, which Internet Archive is one
obvious route to go down that, and to use

350
00:27:36,000 --> 00:27:39,001
the way back machine to go and try and
retrieve some of these journals that have

351
00:27:39,001 --> 00:27:43,000
come and gone. And the second stage is
just to try and link all these things

352
00:27:43,000 --> 00:27:47,001
together. So we don't just have these
digital artifacts that are sitting there

353
00:27:47,001 --> 00:27:51,001
preserved in some way. We can actually
bounce between them and explore those

354
00:27:51,001 --> 00:27:53,001
connections. Thanks very much for your time.

355
00:27:56,000 --> 00:28:01,000
Thank you so much, Rod. And if you would
stay on screen and if Kate would come

356
00:28:01,000 --> 00:28:05,001
back on, I would love to do a quick round
of questions. I do have a question for

357
00:28:05,001 --> 00:28:09,000
each of you. And if again, if there are
additional questions, do please drop them

358
00:28:09,000 --> 00:28:16,000
in the Q&A. But for Kate, the question is,
how much stuff is in Ted Nelson's junk

359
00:28:16,000 --> 00:28:18,001
mail collection and how
did you search through it?

360
00:28:19,000 --> 00:28:23,001
Seems like finding a needle in a haystack.
So actually, I have to, I actually

361
00:28:23,001 --> 00:28:28,001
have to thank Google for that one.
I was just looking for

362
00:28:28,001 --> 00:28:35,001
information on EDP schools.
Actually, I kind of don't, I think it was

363
00:28:35,001 --> 00:28:42,000
just Google. I was looking for ECPI.
I was trying to find information on ECPI and

364
00:28:42,000 --> 00:28:48,001
because the guide had been OCR'd, it showed up.
And I just remember being like,

365
00:28:49,000 --> 00:28:54,001
who's, what is this junk mail collection?
Like what is even happening? And then I

366
00:28:54,001 --> 00:28:59,001
clicked through and I was like, this is a
miracle. Holy cow. I can't believe that

367
00:28:59,001 --> 00:29:04,001
there's this whole thing. And you know,
more than the actual booklet itself,

368
00:29:04,001 --> 00:29:10,000
there were other materials, including
some news articles that had come with the

369
00:29:10,000 --> 00:29:14,001
packet that sort of had encouraged sort
of said like, oh, you know, look how

370
00:29:14,001 --> 00:29:19,000
great, you know, learning to code or,
you know, computer learning computer

371
00:29:19,000 --> 00:29:23,001
programming is and you're going to make
so much money in all of that. So I have

372
00:29:23,001 --> 00:29:29,000
not actually seen enough of, I think,
the junk mail file, but I should

373
00:29:29,000 --> 00:29:31,000
definitely go splunking one day.

374
00:29:31,001 --> 00:29:36,000
That's I bet many of the researchers
here in the audience have the same

375
00:29:36,000 --> 00:29:39,001
experience. I know I did, you know,
doing a Google search for something kind of

376
00:29:39,001 --> 00:29:44,001
obscure in your in your discipline and like,
lo and behold, there it is somewhere

377
00:29:44,001 --> 00:29:48,001
at the Internet Archive. Like I wouldn't
have never thought to go look in the Ted

378
00:29:48,001 --> 00:29:51,000
Nelson junk mail collection for
this thing that I needed. Yeah.

379
00:29:51,001 --> 00:29:57,000
Thank you for that. Kate, a question for you,
Rod from, we'll pick this one up

380
00:29:57,000 --> 00:30:01,001
here from Martin Kaufatevik. Now,
I know that you're not a copyright lawyer, but

381
00:30:01,001 --> 00:30:06,000
I bet you have some opinions on what the
question is that Martin is asking, which

382
00:30:06,000 --> 00:30:10,000
is that current global copyright laws are
terrible for science and for culture,

383
00:30:10,001 --> 00:30:13,000
but people go to jail or
worse for violating them.

384
00:30:13,001 --> 00:30:17,000
Do you have any thoughts on how
institutions can risk manage that?

385
00:30:19,000 --> 00:30:25,000
Well, that's a question. I guess one thing
to point out. So most of the content

386
00:30:25,000 --> 00:30:30,000
that I'm going after, I mean, a lot of it
is it's quote unquote freely available

387
00:30:30,000 --> 00:30:36,000
or some of it is explicitly freely licensed.
So I think there's an element of if

388
00:30:36,000 --> 00:30:40,000
you model is negotiate with the providers
to sign agreements to get the content,

389
00:30:40,000 --> 00:30:43,001
which I guess is the way BHL does,
which is all proper, that mitigates the risk.

390
00:30:45,000 --> 00:30:48,000
I've sort of taken the approach, well,
if it's not kind of like having to

391
00:30:48,000 --> 00:30:51,001
paywall, they're not expressly saying,
do not have this, then we can at least try

392
00:30:51,001 --> 00:30:55,001
and make copies of it so that it
doesn't go away. And again, we've seen

393
00:30:55,001 --> 00:30:59,000
repeatedly, you know,
by the time we could have had these

394
00:30:59,000 --> 00:31:00,001
negotiations,
some of these journals are just gone.

395
00:31:02,000 --> 00:31:07,000
So I take Martin's point, and I guess
that's an interesting question for the

396
00:31:07,000 --> 00:31:10,001
Internet Archive itself.
It's an institution that presumably tries to

397
00:31:10,001 --> 00:31:12,000
mitigate the risk of dealing with this issue.

398
00:31:14,001 --> 00:31:19,000
And Lila Bailey from the policy expert
at the Internet Archive has entered the

399
00:31:19,000 --> 00:31:24,001
chat and dropped off some thoughts,
as did Brewster. So we'll let that

400
00:31:24,001 --> 00:31:28,001
conversation continue on asynchronously.
Kate, Rod, thank you much for your

401
00:31:28,001 --> 00:31:35,001
presentations today. If you have any questions,
drop them into the Q&A, and we'll

402
00:31:35,001 --> 00:31:40,000
pick up with our next speaker.
And so I would like to welcome Michelle

403
00:31:40,000 --> 00:31:46,001
Alexopoulos to the screen today.
And Michelle is going to tell you about her

404
00:31:46,001 --> 00:31:53,001
research looking at expressions on the
federal chairs, the members of the

405
00:31:53,001 --> 00:32:00,001
Board of Federal Reserve Board Chairs,
and how their expressions affect financial

406
00:32:00,001 --> 00:32:02,001
markets. So over to you, Michelle.

407
00:32:03,001 --> 00:32:09,001
Thank you very much. I would like to thank
the organizers, obviously, for putting

408
00:32:09,001 --> 00:32:16,001
together such an interesting and eclectic
group of papers. So as Christa said,

409
00:32:16,001 --> 00:32:20,001
I want to talk about more than words.
Basically, this is an analysis of the

410
00:32:20,001 --> 00:32:25,001
Federal Reserve Board Chairs' communication
during congressional testimony based

411
00:32:25,001 --> 00:32:29,000
on a project that I'm doing with members
of Bank of Canada, specifically Zhen

412
00:32:29,000 --> 00:32:33,000
Tanhan, Alaski, Prashof, and Xu Zhang.
The link to our newly released working

413
00:32:33,000 --> 00:32:37,001
paper is available there.
Now, in terms of the product data sources, this

414
00:32:37,001 --> 00:32:41,001
wouldn't have happened unless we had access
to a tremendous amount of resources.

415
00:32:42,000 --> 00:32:46,000
We're looking for audio, video,
as well as textual. From the Internet Archive,

416
00:32:46,001 --> 00:32:51,001
we've primarily used the TV Archive content.
These are videos from C-SPAN, CNBC,

417
00:32:52,000 --> 00:32:56,001
Bloomberg TV, and the House and Senate videos.
And when there's polls in that

418
00:32:56,001 --> 00:33:02,001
coverage, we've then turned to C-SPAN archives,
YouTube videos, textual

419
00:33:02,001 --> 00:33:07,001
sources that would give us transcript
information. And what we did is we analyzed

420
00:33:07,001 --> 00:33:11,001
all of this and then tried to blend it
with basically tick by tick stock data

421
00:33:11,001 --> 00:33:14,001
that came from places like
Wharton and Refititive, etc.

422
00:33:15,001 --> 00:33:19,000
So the overview of the project is pretty
straightforward. What we wanted to do is

423
00:33:19,000 --> 00:33:23,000
examine the importance and the impact of
the US Federal Reserve communication.

424
00:33:23,001 --> 00:33:27,001
We're going to measure aspects of the
communication using AI, natural language

425
00:33:27,001 --> 00:33:31,001
processing, and other tools.
And then we want to test if and how the words

426
00:33:31,001 --> 00:33:34,001
chosen, their body language,
and their tone of voice is actually affecting

427
00:33:34,001 --> 00:33:39,000
markets. So when we think about policy
communication, I understand we have a very

428
00:33:39,000 --> 00:33:44,000
large audience. When I talk about monetary
policy and policy communications, I'm

429
00:33:44,000 --> 00:33:48,000
talking about things such as interest rate
movements for the policy rate, forward

430
00:33:48,000 --> 00:33:52,001
guidance, quantitative easing,
or bank balance sheet operations. These things

431
00:33:52,001 --> 00:33:56,001
have very large impacts on the welfare
of businesses and individuals in our

432
00:33:56,001 --> 00:34:01,000
economy, not just in the United States,
but because they're so large around the

433
00:34:01,000 --> 00:34:05,000
world. Now, these policies are often
fairly complex. Now, to be effective,

434
00:34:05,001 --> 00:34:09,000
central bank communications need to be
accurate. They need to be clear. They need

435
00:34:09,000 --> 00:34:12,001
to be perceived as credible, and they
need to reach their target audience. And

436
00:34:12,001 --> 00:34:17,000
it's true that when the central bank takes
actions or releases communications, we

437
00:34:17,000 --> 00:34:21,001
all are not necessarily watching the
actual release, but we're also getting our

438
00:34:21,001 --> 00:34:25,001
information through media. So we have
two different types of channels where we

439
00:34:25,001 --> 00:34:28,001
get this information. It could be
direct communication, such as the press

440
00:34:28,001 --> 00:34:33,000
conferences themselves, or through watching
things like the testimony, or even

441
00:34:33,000 --> 00:34:38,000
the press releases, such as the FOMC
statements that were released last week, or

442
00:34:38,000 --> 00:34:41,000
we can have the indirect types of
communications that we often see captured by

443
00:34:41,000 --> 00:34:47,001
the Internet Archives TV archive,
which deals with the coverage on CNBC, Fox

444
00:34:47,001 --> 00:34:53,000
News, or even things like Bloomberg TV.
This is also then amplified by sort of

445
00:34:53,000 --> 00:34:57,000
traditional things such as the New York
Times and Twitter. Now, when we think

446
00:34:57,000 --> 00:35:01,001
about communication, communication is
obviously more than words. And this has

447
00:35:01,001 --> 00:35:05,001
been popularized for a very long time,
ever since the 1970s. I'm sure some of you

448
00:35:05,001 --> 00:35:10,000
have heard that the statement that over 90%
of communication it comes from your

449
00:35:10,000 --> 00:35:14,001
body language and tone of voice.
Now, although the Moravian studies are probably

450
00:35:14,001 --> 00:35:19,001
not getting exactly the right numbers in
terms of which quantities, the idea is

451
00:35:19,001 --> 00:35:25,000
pretty much accepted. So you can see here
even a few pictures of Ben Bernanke and

452
00:35:25,000 --> 00:35:28,000
Janet Yellen, who are two of the
Federal Reserve chairs, and you can see

453
00:35:28,000 --> 00:35:32,001
sometimes they look very relaxed,
and sometimes they look a lot more perplexed or

454
00:35:32,001 --> 00:35:37,001
concerned. So we want to think about how
that actually influences people. And why

455
00:35:37,001 --> 00:35:41,001
does body language potentially matter?
Body language and tone of voice could

456
00:35:41,001 --> 00:35:46,000
matter through a direct channel.
So a trader or us individuals may be watching

457
00:35:46,000 --> 00:35:49,001
directly, and then we take actions based
on what it is we're hearing. We're

458
00:35:49,001 --> 00:35:53,001
either actively looking for something,
or it may affect how we feel about

459
00:35:53,001 --> 00:35:57,001
something or sentiment or confidence.
If we're looking at indirect channels,

460
00:35:57,001 --> 00:36:02,001
because these get picked up and amplified
through other TV stations and other

461
00:36:02,001 --> 00:36:06,000
news outlets,
what we can basically get is that the Fed has a

462
00:36:06,000 --> 00:36:08,001
communication,
the journalists or the analysts are

463
00:36:08,001 --> 00:36:10,000
watching it,
and then they're releasing information.

464
00:36:10,001 --> 00:36:13,001
And now those of us who are receiving
that information, it could be broken

465
00:36:13,001 --> 00:36:18,000
telephone, may not be, but we'll take
actions upon that themselves. So you can

466
00:36:18,000 --> 00:36:23,000
see here, for example, a quote that came
from the CNBC street signs that was

467
00:36:23,000 --> 00:36:27,000
talking about when Janet Yellen looked
insecure, when she looked frustrated, when

468
00:36:27,000 --> 00:36:31,001
she looked angry, and even when Ben
Bernanke's voice potentially got shaky at

469
00:36:31,001 --> 00:36:36,001
various times. And then there's also the
impact of algorithmic traders, where now

470
00:36:36,001 --> 00:36:39,001
we have an automated channel where they
may be taking a lot of these signals,

471
00:36:39,001 --> 00:36:43,001
analyzing them and trading on their behalf.
So what our research tries to do is

472
00:36:43,001 --> 00:36:47,000
look at these different channels,
we're not going to take a stand on which one is

473
00:36:47,000 --> 00:36:50,001
most important. But we want to take a look
at Fed chair testimonies. And this is

474
00:36:50,001 --> 00:36:54,000
a little different than some of the other
papers that have been looking at Fed

475
00:36:54,000 --> 00:36:58,000
communications, because we're focusing on
these 32 testimonies, which will expand

476
00:36:58,000 --> 00:37:01,001
over the summer. And we're looking
at the three different dimensions of

477
00:37:01,001 --> 00:37:05,001
communications all at once. Now, what's
interesting about these communications is

478
00:37:05,001 --> 00:37:10,000
there's also a prepared portion of this.
And we have unscripted question and

479
00:37:10,000 --> 00:37:14,000
answers. And people can obviously respond
to those things in different ways. We

480
00:37:14,000 --> 00:37:17,001
will also have the responses from
the senators and the congressional

481
00:37:17,001 --> 00:37:22,001
representatives. Now, why testimonies,
you might ask, well, first of all, they're

482
00:37:22,001 --> 00:37:26,000
widely watched by Fed watchers and investors.
It's

483
00:37:26,000 --> 00:37:27,001
covered by a lot of news media.

484
00:37:28,000 --> 00:37:31,001
It doesn't occur on the same day as an
actual policy announcement. So you can

485
00:37:31,001 --> 00:37:35,001
focus a little bit more on disentangling
just the communications component. And

486
00:37:35,001 --> 00:37:38,001
of course, the high quality TV footage
and transcripts that are available from

487
00:37:38,001 --> 00:37:43,001
the archives has made this actually
a doable thing at the moment. So the

488
00:37:43,001 --> 00:37:46,000
structure of the semi annual testimony
is these things are about two to three

489
00:37:46,000 --> 00:37:50,000
hours, and they come in pairs.
So the semi annual testimonies, you end up with

490
00:37:50,000 --> 00:37:54,001
one before the house and one before the
Senate Banking Committee, you have the

491
00:37:54,001 --> 00:37:58,001
prepared remarks released before things begin,
then you'll have some welcoming

492
00:37:58,001 --> 00:38:02,001
statements, the prepared remarks in the Q&A.
What we can do then is we can

493
00:38:02,001 --> 00:38:06,001
disentangle the different components here,
we have transcripts, which are going

494
00:38:06,001 --> 00:38:10,001
to give us the actual words that are spoken,
we have the audio component, and

495
00:38:10,001 --> 00:38:14,000
then we have the video component,
which will allow us to look at how facial

496
00:38:14,000 --> 00:38:18,001
expressions are formed. So here,
the three different ways, basically from the

497
00:38:18,001 --> 00:38:22,000
text, without getting into details,
we're using a fine trained BERT model, which

498
00:38:22,000 --> 00:38:26,000
is a language model that sort of state of
the art right now from the voice, we're

499
00:38:26,000 --> 00:38:31,000
basically forcing alignments with transcripts,
and we're extracting pitch for

500
00:38:31,000 --> 00:38:37,000
each speaker, with a tool called Pratt.
And for the facial expressions, we're

501
00:38:37,000 --> 00:38:41,000
using things that would be used in
laboratories and psychology. So face reader,

502
00:38:41,001 --> 00:38:46,000
for example, we're using Microsoft Azure,
and we're matching these to look at

503
00:38:46,000 --> 00:38:50,000
facial action units, and then using
mappings that have been done with

504
00:38:50,000 --> 00:38:54,001
psychologists before, into figuring out
what kind of negative emotions might be

505
00:38:54,001 --> 00:38:58,000
being expressed by the Federal Reserve
Chairs during this time. Now, all in all,

506
00:38:58,000 --> 00:39:01,001
we take all of this, we then have to mash
it in with financial market data, we

507
00:39:01,001 --> 00:39:03,001
have to be very careful
about the timing of this.

508
00:39:03,001 --> 00:39:09,000
So we use the CNBC live coverage,
and we're watching what the values of the S&P

509
00:39:09,000 --> 00:39:14,001
500 are, in order to do some of that
crosswalk in the timestamps. Then what we do

510
00:39:14,001 --> 00:39:18,001
is we basically create this giant database,
we're going to use local projection

511
00:39:18,001 --> 00:39:23,001
regressions and difference in different
type projections, in order to look at the

512
00:39:23,001 --> 00:39:27,001
impact of these different channels of
emotions on the outcomes, which would be

513
00:39:27,001 --> 00:39:31,000
say the change in the S&P 500,
or what's happening with the VIX.

514
00:39:32,000 --> 00:39:35,000
Now, what we typically find,
and this is just one of the examples, is yes,

515
00:39:35,000 --> 00:39:39,001
indeed, the S&P 500 and the VIX tend to
respond in ways that you might expect. So

516
00:39:39,001 --> 00:39:45,000
if they're actually relaxed and happy,
not negative, you end up seeing the values

517
00:39:45,000 --> 00:39:49,000
of the stock markets going up in the short run,
and the value of the VIX, which

518
00:39:49,000 --> 00:39:51,001
is usually considered to be
a fear index going down.

519
00:39:52,001 --> 00:39:55,001
Now, what you also see is there are
differences across different topics, and as

520
00:39:55,001 --> 00:39:59,000
you might expect, the ones that the Fed
has most oversight or control on, such as

521
00:39:59,000 --> 00:40:03,000
monetary policy, will have the largest
impact on markets over the time period.

522
00:40:04,000 --> 00:40:06,001
Now, the summaries of the findings today,
because I did promise Chris, I would

523
00:40:06,001 --> 00:40:10,001
try to get in on time, is what we did
find is these different dimensions, they

524
00:40:10,001 --> 00:40:14,000
don't all act at the same time in a highly
correlated fashion. Text, voice, and

525
00:40:14,000 --> 00:40:18,001
facial emotions, though, do tend to move
financial markets. I showed you a couple

526
00:40:18,001 --> 00:40:22,000
of examples for the short run responses,
but the impact seems to grow stronger

527
00:40:22,000 --> 00:40:23,001
in the days following these testimonies.

528
00:40:24,000 --> 00:40:27,001
There's differences in the amount of soft
information, how they can differ across

529
00:40:27,001 --> 00:40:31,000
topics, and we also are starting to see
some patterns that we're going to be

530
00:40:31,000 --> 00:40:34,001
verifying over the summer as to whether
or not some of these responses differ

531
00:40:34,001 --> 00:40:39,001
from different Fed chairs. We also have
congressional members' emotions, and

532
00:40:39,001 --> 00:40:42,001
sometimes they will significantly
move markets as well.

533
00:40:43,001 --> 00:40:47,000
All right, so with that, I say thank you.
Comments and questions are welcome.

534
00:40:47,000 --> 00:40:51,000
Again, there's a link to a working paper,
and you can reach out to me via email.

535
00:40:51,000 --> 00:40:56,001
I'll be happy to have a conversation.
Thank you very much. Thank you, Michelle.

536
00:40:57,000 --> 00:41:00,001
We have gotten a couple of questions that
have come in to me. Again, if you have

537
00:41:00,001 --> 00:41:04,000
additional questions for Michelle,
please drop them off in the Q&A. But what we'd

538
00:41:04,000 --> 00:41:09,001
like to do now is move on to our next speaker.
Again, we'll pick up questions

539
00:41:09,001 --> 00:41:15,001
with both Sawood and Michelle in just a minute.
But today, right now, I want to

540
00:41:15,001 --> 00:41:19,001
welcome Sawood along to the screen.
Sawood works for the Internet Archive. He's

541
00:41:19,001 --> 00:41:25,001
part of the Wayback Machine team doing
amazing things. He's going to walk you

542
00:41:25,001 --> 00:41:30,000
through some of the work that he's doing
on metadata behind the scenes with our

543
00:41:30,000 --> 00:41:33,000
web archive. Sawood, please, over to you.

544
00:41:33,001 --> 00:41:36,001
Hi. Allow me to share my screen.

545
00:41:40,000 --> 00:41:46,001
Taking a while. Okay, that's top one.

546
00:41:50,001 --> 00:41:55,000
Is my screen visible? Looks great.
Awesome. Thank you.

547
00:41:55,001 --> 00:41:58,000
[...] Sawood Alam, a web and data scientist of

548
00:41:58,000 --> 00:42:00,000
Wayback Machine at the Internet Archive.

549
00:42:01,000 --> 00:42:06,000
What you see on my screen is the help
text of a tool called CDX summaries.

550
00:42:07,000 --> 00:42:12,001
In my current working directory,
I have a file called sample.cdx.gc, which is a

551
00:42:12,001 --> 00:42:18,001
compressed index of a bunch of web archive
files. But we don't know what was in

552
00:42:18,001 --> 00:42:24,001
those web archive files or work files.
Hopefully, when we run this index through

553
00:42:24,001 --> 00:42:28,001
our CDX summary tool, we'll get to know
what is inside of there. I'm going to run

554
00:42:28,001 --> 00:42:33,000
this command here, and it will take a while
to complete. So we'll come back to it

555
00:42:33,000 --> 00:42:39,000
later. Let's go to the Internet Archive
homepage. Here we see a bunch of

556
00:42:39,000 --> 00:42:44,001
collections listed. Some are collections
of books and audio and video. Some are

557
00:42:44,001 --> 00:42:50,001
web collections. Let's open one of
these collections. In this case, it is

558
00:42:50,001 --> 00:42:56,000
Ukrainian cultural heritage collection.
You scroll through this collection. You

559
00:42:56,000 --> 00:43:02,000
see some interesting colorful thumbnails,
some meaningful titles, and you have a

560
00:43:02,000 --> 00:43:08,000
rough idea of what to see or to get in
this collection. Now, let's go to a web

561
00:43:08,000 --> 00:43:13,001
collection, which is a collection of WAP files.
An arbitrary number of web

562
00:43:13,001 --> 00:43:19,001
captures are packaged together. And in
this case, we see lists of tombstones here

563
00:43:19,001 --> 00:43:26,000
as thumbnails. And titles look very similar.
I guess they were generated by

564
00:43:26,000 --> 00:43:31,001
machines. It doesn't tell a lot.
There is an about tab in this collection view,

565
00:43:31,001 --> 00:43:33,000
but we don't see much.

566
00:43:33,001 --> 00:43:38,000
There are a few lines of metadata here
and that's all. And this is the problem

567
00:43:38,000 --> 00:43:44,001
that you are going to solve here.
In the next few days or weeks, when the testing

568
00:43:44,001 --> 00:43:50,000
is done and the changes are merged to
the main site, this page will look

569
00:43:50,000 --> 00:43:57,000
something like this.
We'll get to know how many captures are in this

570
00:43:57,000 --> 00:44:02,001
collection, how many unique URIs were captured,
how many of those URIs belong to,

571
00:44:02,001 --> 00:44:09,001
say, SGMN pages in 200 okay answers,
for example, or how many of those URLs have

572
00:44:09,001 --> 00:44:16,000
zero paths and zero queries that is root
pages versus deep links. We can also get

573
00:44:16,000 --> 00:44:17,001
a total spread of the collection.

574
00:44:18,000 --> 00:44:24,001
So this collection started in 2016 and
stopped there. No activity in 2017

575
00:44:24,001 --> 00:44:30,000
and then started again 2018. And since then,
it has been constantly archiving

576
00:44:30,000 --> 00:44:35,000
every single month. Then we also have a
bunch of top posts that are contributing

577
00:44:35,000 --> 00:44:37,001
to this collection with
maximum number of captures.

578
00:44:38,000 --> 00:44:45,000
And one cool thing that the tool brings
up is a bunch of sample URIs from this

579
00:44:45,000 --> 00:44:51,000
collection. So WAC files are bundled up a
number of URIs that you don't know what

580
00:44:51,000 --> 00:44:55,001
is inside in there. And this allows us to
click on one of these links and see how

581
00:44:55,001 --> 00:45:00,000
well those pages were archived.
So it helps us doing QA, for example. All these

582
00:45:00,000 --> 00:45:06,000
numbers appear. They also allow us to
gain insights and learn sometimes if the

583
00:45:06,000 --> 00:45:10,000
numbers are really off, we learn like
there was some problem in our calling

584
00:45:10,000 --> 00:45:14,001
process and go back and fix it. So how
this whole page was rendered is basically

585
00:45:14,001 --> 00:45:20,001
backed by a JSON file. And this JSON file
is generated using a command line tool

586
00:45:20,001 --> 00:45:25,001
called CDX summaries. And we have open
sourced this tool. So you can use this

587
00:45:25,001 --> 00:45:32,000
tool to generate human readable summaries
or machine readable JSON files. And if

588
00:45:32,000 --> 00:45:37,001
you have a JSON file, you can render it
in HTML using this web component that we

589
00:45:37,001 --> 00:45:42,001
made available. So with that,
let's go back to the command line and see what

590
00:45:42,001 --> 00:45:48,000
happened to the local CDX file that we had.
And it turned out it did generate a

591
00:45:48,000 --> 00:45:54,000
very nice human readable summary of the
CDX index that we had. Now we get to know

592
00:45:54,000 --> 00:45:58,000
what is inside in there, when it was captured,
and how many captures are there,

593
00:45:58,000 --> 00:46:03,001
and so on and so forth. And we can even
click on one of these links to open them

594
00:46:03,001 --> 00:46:08,000
in a browser and see how it was archived.
But by demonstrating it on command

595
00:46:08,000 --> 00:46:12,001
line, what we really illustrated is this
tool is not tied to internet archives.

596
00:46:13,001 --> 00:46:17,000
It is an independent tool, actually.
We are just using an internet archive.

597
00:46:17,001 --> 00:46:22,000
Anyone can use it in their own web
archival collection. So with that, I will

598
00:46:22,000 --> 00:46:29,000
recap. We basically created a command
line tool and a companion web component

599
00:46:29,000 --> 00:46:35,000
to summarize the web collections.
We released these tools under open source

600
00:46:35,000 --> 00:46:40,001
license. And we use these tools to
enrich our own web collections.

601
00:46:41,001 --> 00:46:47,001
We gain insight from generated stats
to improve our color. Finally, we

602
00:46:47,001 --> 00:46:52,000
demonstrated that these tools are not
exclusive to the internet archive. So

603
00:46:52,000 --> 00:46:56,001
anyone can use them to their own web
archival collection and learn more about

604
00:46:56,001 --> 00:47:01,001
what they are having. With that,
I will see you in QA. And thank you so much for

605
00:47:01,001 --> 00:47:07,000
listening. Thank you so much.
So we would, you know, lots of comments and

606
00:47:07,000 --> 00:47:12,000
questions coming in in the chat.
So I think we will have a good question for you

607
00:47:12,000 --> 00:47:18,001
in just a minute. But up next,
we have the final bit of this segment is a

608
00:47:18,001 --> 00:47:23,001
presentation, a video from Art Rhino.
And I had the chance of watching this

609
00:47:23,001 --> 00:47:27,000
video. And I think it is going to blow
your mind. It is a little Willy Wonka and

610
00:47:27,000 --> 00:47:33,000
just charming and wonderful.
So Art is going to show you in this video kind

611
00:47:33,000 --> 00:47:39,001
of the internet archives scanning frames,
the TT scanner.

612
00:47:40,000 --> 00:47:42,000
So, Caitlin, let's roll the video.

613
00:47:45,001 --> 00:47:48,001
Hi, there. My name is Art Rhino.
I work at the University of

614
00:47:48,001 --> 00:47:50,000
Windsor in Ontario, Canada.

615
00:47:51,000 --> 00:47:55,001
And I'm on the board of a nonprofit
organization called our Digital World or ODW.

616
00:47:56,000 --> 00:48:00,000
My library acquired a very nice
tabletop scribe scanning machine from

617
00:48:00,000 --> 00:48:01,001
the internet archive years ago.

618
00:48:01,001 --> 00:48:06,000
This unit currently has good availability.
And I've become interested in scanning

619
00:48:06,000 --> 00:48:10,000
the back file for a major paper collection.
A lot of the material is well

620
00:48:10,000 --> 00:48:14,001
positioned for scanning, some of it can
be disassembled and the pages can be fed

621
00:48:14,001 --> 00:48:19,001
through our Xerox scan feature.
However, there's a lot of stitch binding. I've

622
00:48:19,001 --> 00:48:23,001
been trying to find a way to scan it
without compromising the binding. Here are

623
00:48:23,001 --> 00:48:27,000
the components I've used to pull together
automatic scanning, most of which were

624
00:48:27,000 --> 00:48:33,001
already on hand from ODW's work on DIY
microfilm scanners. When new addition is

625
00:48:33,001 --> 00:48:38,001
an inflatable snow tube, which came
from the end of season table at a local

626
00:48:38,001 --> 00:48:43,001
hardware store, the air mattress pump
inflates the snow tube and a vacuum cleaner

627
00:48:43,001 --> 00:48:48,000
holds on to the page. The device in the
front that looks like a car is the MBOT

628
00:48:48,000 --> 00:48:54,000
or M-Block MBOT. My MBOT is several years old.
And I use the M-Block ID on the

629
00:48:54,000 --> 00:48:58,001
web for controlling it. There are wireless
and Bluetooth options for newer models

630
00:48:58,001 --> 00:49:02,001
that would be more elegant for this kind
of thing. However, this worked well

631
00:49:02,001 --> 00:49:06,001
enough and it seems to do the trick for
what we want to do. Probably the biggest

632
00:49:06,001 --> 00:49:10,001
drawback of all this is the noise from
the air mattress pump and the vacuum

633
00:49:10,001 --> 00:49:15,000
cleaner, which you can't hear here.
But otherwise, I'm pretty happy with this

634
00:49:15,000 --> 00:49:19,000
approach. It takes about 30 to
45 minutes to scan the typically

635
00:49:19,000 --> 00:49:20,001
100 pages in the title.

636
00:49:20,001 --> 00:49:25,001
You can find way more sophisticated and
faster devices on YouTube. And it would

637
00:49:25,001 --> 00:49:29,001
be cool to make something more general
purpose and robust. But this is one

638
00:49:29,001 --> 00:49:33,001
approach for automatically scanning
material on the tabletop scribe that is

639
00:49:33,001 --> 00:49:36,001
hopefully useful for others.
Thanks for listening.

640
00:49:39,001 --> 00:49:44,000
So I have it on good authority that
art does a lot of really interesting

641
00:49:44,000 --> 00:49:50,001
projects. And as you can see,
as a little bit of a tinkerer and just what

642
00:49:50,001 --> 00:49:54,001
creativity. I think that someone mentioned
like a rotisserie chicken turning

643
00:49:54,001 --> 00:50:00,001
thing next to a microfilm scanner.
Anyway, interesting stuff there.

644
00:50:01,000 --> 00:50:06,001
I would like to welcome or I'd like to
bring Michelle and Sowood back to the

645
00:50:06,001 --> 00:50:11,001
screen for a couple of questions.
And so the first question that I have, if you

646
00:50:11,001 --> 00:50:17,001
want to turn your cameras on, Michelle,
the question that I have is for you. Are

647
00:50:17,001 --> 00:50:21,001
there data points or related data that
you'd like to include in your research

648
00:50:21,001 --> 00:50:26,000
that simply weren't or aren't available
in digital or computable form?

649
00:50:27,000 --> 00:50:32,001
So that's a great question. There are holes
in the collection. Unfortunately, the

650
00:50:32,001 --> 00:50:37,001
Internet Archive hasn't gone back quite
far enough for what we'd love to do,

651
00:50:37,001 --> 00:50:42,001
which is kind of go back sort of to the
beginning to include a lot more of the

652
00:50:42,001 --> 00:50:49,000
data from, say, Greenspan's time
period or even people passed on

653
00:50:49,000 --> 00:50:54,000
that. And sometimes some of the
audio files and some other forms of

654
00:50:54,000 --> 00:50:58,000
communications that could have happened
around the time. So some of that's not

655
00:50:58,000 --> 00:51:02,001
available. Some we're still trying to dig
through some archives to see whether or

656
00:51:02,001 --> 00:51:07,001
not they've been put someplace else or
sort of misplaced on some of that. But

657
00:51:07,001 --> 00:51:11,001
there is the unfortunate thing that I mean,
why the Internet Archive is so

658
00:51:11,001 --> 00:51:16,000
important that things have gotten destroyed
over time. So we would really like to

659
00:51:16,000 --> 00:51:20,001
have it as inclusive as possible.
We'd also like to look at some of the issues

660
00:51:20,001 --> 00:51:25,000
between Janet Yellen and other female Fed
authorities and whether or not there's

661
00:51:25,000 --> 00:51:29,001
different responses on them.
And it's just sometimes the video feeds and

662
00:51:29,001 --> 00:51:33,001
everything like that just aren't there
right now. So those would be the primary

663
00:51:33,001 --> 00:51:37,000
ones. We're getting much better at getting
stock data and everything coming in.

664
00:51:37,001 --> 00:51:42,000
So that's not quite so much of a problem.
Fascinating. Thanks so much for that.

665
00:51:42,001 --> 00:51:46,000
So, Wud, a question for you.
The comment is, thank you for making

666
00:51:46,000 --> 00:51:48,000
your CDX tool open source.

667
00:51:49,000 --> 00:51:52,000
What other projects would you
envision using your tool?

668
00:51:54,000 --> 00:52:00,000
Ah, interesting. So one immediate
application I was thinking the other day is

669
00:52:00,000 --> 00:52:06,001
there is an emerging format called Wack-Z,
which is like

670
00:52:06,001 --> 00:52:11,001
bundling a bunch of WACK files with their
indices and all that stuff. I think

671
00:52:11,001 --> 00:52:16,000
this tool can integrate well inside in there.
So if someone kind of loads a Wack

672
00:52:16,000 --> 00:52:21,000
-Z file in a browser, they can see what
really is inside. And it has been a

673
00:52:21,000 --> 00:52:26,001
challenge like I mean, Mark Graham,
director of the Bayback Machine, he often

674
00:52:26,001 --> 00:52:31,001
asked me, okay, so here is this item we
collected, what is inside in there? And

675
00:52:31,001 --> 00:52:35,001
you know, it's a tool go around against it.
And you will have like a bunch of

676
00:52:35,001 --> 00:52:41,000
URLs that you can play with and explore
from there. Again, I mean, this is both

677
00:52:42,001 --> 00:52:47,001
statistical insights, as well as kind of
exploratory. Another option I've been

678
00:52:47,001 --> 00:52:54,001
looking forward for is to kind of
know these random URLs that we pull

679
00:52:54,001 --> 00:53:00,000
out from these. And those are not just random,
they have some other criteria to

680
00:53:00,000 --> 00:53:05,001
that. We can use those to generate
thumbnails and have like a slideshow or

681
00:53:05,001 --> 00:53:11,001
something like that to have a more visual
understanding of what's inside in a

682
00:53:11,001 --> 00:53:16,000
collection and use it as a way to kind
of tell a story about a collection.

683
00:53:17,001 --> 00:53:23,001
That's bad. Thank you for that explanation.
I understand from a message from

684
00:53:23,001 --> 00:53:26,001
Michael Nelson in chat that you're going
to be teaching a course next semester.

685
00:53:27,001 --> 00:53:32,001
Oh, yes, it will be a web server design.
So yeah, my students will be competing

686
00:53:32,001 --> 00:53:38,001
with the engine X and Apache and some of
other fancy web servers, hoping to have

687
00:53:38,001 --> 00:53:45,001
a more, you know,
compliant web server with RFC or standard then

688
00:53:45,001 --> 00:53:52,001
some of these other well known web server.
Well, good luck

689
00:53:52,001 --> 00:53:58,001
to you in the in the course and your
students are in excellent hands for sure.

690
00:53:59,000 --> 00:54:03,001
Michelle, so we thank you so much for your
presentations today. And we'd like to

691
00:54:03,001 --> 00:54:07,001
keep moving through the through today's talks.
If you had additional questions

692
00:54:07,001 --> 00:54:14,000
for Michelle, or so would please do
drop them off into the Q&A. But what

693
00:54:14,000 --> 00:54:20,000
Spencer Torin to the to the screen from
Thompson Reuters Special Services and

694
00:54:20,000 --> 00:54:24,000
Spencer is going to talk about the work
that he's done with automatic hashtag

695
00:54:24,000 --> 00:54:29,000
hierarchy generation. And I can't wait
to hear that. So Spencer, over to you.

696
00:54:32,000 --> 00:54:38,001
Yep. Okay, hello.

697
00:54:39,000 --> 00:54:45,001
Welcome, everyone. Yeah, so I'd like to
talk about, well, what Chris just said.

698
00:54:46,000 --> 00:54:51,001
So the idea is, you know, we have all
these tags hashtags on Twitter. And the

699
00:54:51,001 --> 00:54:57,000
question is, do they have a hierarchy is
their structure to them. And I'll talk a

700
00:54:57,000 --> 00:55:01,000
little bit about the difference between
ontologies and folks on amis really

701
00:55:01,000 --> 00:55:06,000
quickly. So we're all probably if not
familiar with ontologies by name, at least

702
00:55:06,000 --> 00:55:11,000
by concept. So we have things like
the dictionary, shall we say, in the

703
00:55:11,000 --> 00:55:15,001
dictionary, there are very curated terms,
there are rigorous definitions of these

704
00:55:15,001 --> 00:55:19,000
terms. And there are, you know, well defined
relations between these terms. So an

705
00:55:19,000 --> 00:55:23,001
example is a foot has five toes and is
part of the human body, there are many

706
00:55:23,001 --> 00:55:27,001
more definitions of what a foot is,
you can see on the right, and I think on the

707
00:55:27,001 --> 00:55:32,000
right is just some of them. But we have
to compare that with the difficulty of

708
00:55:32,000 --> 00:55:39,000
looking at, let's say, Twitter, in folks

709
00:55:39,000 --> 00:55:45,000
on ami. So these hashtags that are used
are very arbitrary. Their definitions are

710
00:55:45,000 --> 00:55:52,000
circumstantial. There are very undefined
relationships to them. I just created a

711
00:55:52,000 --> 00:55:55,001
hashtag down there to hashtags down there.
And if I use them on Twitter,

712
00:55:55,001 --> 00:55:57,001
they would now exist in the record.

713
00:55:58,001 --> 00:56:03,000
What do some of those hashtags even mean?
I don't even know, really, I just put

714
00:56:03,000 --> 00:56:08,000
up some nonsense just to illustrate the
fact that I could do it. So, you know, if

715
00:56:08,000 --> 00:56:13,000
we want to put some structure to these
hashtags, we have to be able to ask at

716
00:56:13,000 --> 00:56:16,000
least the minimal question, which is,
you know, are there hashtags which are more

717
00:56:16,000 --> 00:56:21,001
general than others, shall we say?
Is hashtag dog more general than Dalmatian? If

718
00:56:21,001 --> 00:56:26,000
dog in our in our ontological thinking
is more general than Dalmatian, is that

719
00:56:26,000 --> 00:56:30,000
true on Twitter? Is AI hashtag AI more
general than hashtag machine learning?

720
00:56:30,001 --> 00:56:37,000
What does general mean on Twitter?
An ontology would say that a general word

721
00:56:37,000 --> 00:56:42,000
is perhaps due to the breadth of applicability
of the term. And that again would

722
00:56:42,000 --> 00:56:47,001
be by the definition. There's nothing to
say that you could or should use a term

723
00:56:47,001 --> 00:56:52,001
very often. But by the definition of the term,
you could say that the term is

724
00:56:52,001 --> 00:56:58,000
general. With folks on amis, since there's
no real clear definition for these

725
00:56:58,000 --> 00:57:02,000
things, we probably have to take the
question in a different direction and talk

726
00:57:02,000 --> 00:57:06,000
about the breadth of application rather
than the breadth of applicability. So,

727
00:57:06,000 --> 00:57:09,000
we're going to ask certain contextual
questions, like, you know, when was it

728
00:57:09,000 --> 00:57:15,001
used? Where was it used? Who used it?
How was it used? So, we look at contexts as

729
00:57:15,001 --> 00:57:19,000
proxies for generality.
So, there's a classic one,

730
00:57:19,000 --> 00:57:20,001
which would be the popularity context.

731
00:57:21,000 --> 00:57:25,000
If a lot of people are using it,
that might indicate it's for general use. There

732
00:57:25,000 --> 00:57:31,001
are lots of contexts. Here are examples.
Hashtag disease would be not

733
00:57:31,001 --> 00:57:36,001
related to a specific event, although
COVID-19 is. It's a big deal. It's a big

734
00:57:36,001 --> 00:57:42,000
event. But it's nonetheless an event.
And in some years, probably we'll see that

735
00:57:42,000 --> 00:57:47,000
hashtag die out. You can have a general
holiday, which could happen any time of

736
00:57:47,000 --> 00:57:51,000
the year. Halloween is generally one time
of year. You could have hashtag work,

737
00:57:51,001 --> 00:57:55,000
which occurs throughout the week.
TGIF might be a Friday thing. And you could

738
00:57:55,000 --> 00:58:00,001
certainly have the hashtag food any time
of the day and breakfast maybe more so

739
00:58:00,001 --> 00:58:06,000
in the morning. And then the all-important
semantic contexts. So, you know, if

740
00:58:06,000 --> 00:58:11,001
you have a hashtag that co-occurs with
many other different hashtags, this is

741
00:58:11,001 --> 00:58:15,000
sort of the state of the art of the moment,
you know, which other, how many other

742
00:58:15,000 --> 00:58:20,000
different unique hashtags is a
hashtag used with. So, if very

743
00:58:20,000 --> 00:58:21,001
many, then we could call it general.

744
00:58:21,001 --> 00:58:25,000
And if not very many, then maybe not so
general. And then we can also look at

745
00:58:25,000 --> 00:58:29,000
topics, which you might consider to
be groups of hashtags. Is it used

746
00:58:29,000 --> 00:58:30,001
within a group or outside the group?

747
00:58:31,001 --> 00:58:38,001
And also tokens and words. So, is it used
with many different ideas or not? And

748
00:58:38,001 --> 00:58:42,000
the way we measure all this is basically
we call it the, well, we don't call it

749
00:58:42,000 --> 00:58:45,001
the Shannon diversity index is what
it is called. This is an ecologically

750
00:58:45,001 --> 00:58:51,001
inspired term. It is just Shannon entropy.
For those of you who are not familiar

751
00:58:51,001 --> 00:58:56,000
with Shannon entropy, I will briefly
describe it. So, on the left, you see

752
00:58:56,000 --> 00:59:01,000
something with very low entropy, very low
diversity. You kind of know what you're

753
00:59:01,000 --> 00:59:05,000
getting with that thing, where it fits,
what context it's used in. And then over

754
00:59:05,000 --> 00:59:09,000
on the very far right, you would have no
idea when and where you might see this

755
00:59:09,000 --> 00:59:14,001
thing. So, it would be, you might consider
it to be very diverse. And then let me

756
00:59:14,001 --> 00:59:17,001
just take, you know, I mentioned the
eight different contexts a few slides

757
00:59:17,001 --> 00:59:22,000
previously. That's what, you know,
you have eight of these diversity measures for

758
00:59:22,000 --> 00:59:26,000
each hashtag. And then you just multiply
them by some weighting and you come up

759
00:59:26,000 --> 00:59:31,000
with this ensemble diversity index.
And that value would be an indication of how

760
00:59:31,000 --> 00:59:35,001
general that hashtag is. The higher the value,
the more general it is. The lower

761
00:59:35,001 --> 00:59:38,000
the value, the less general it is,
the more specific.

762
00:59:39,000 --> 00:59:41,001
So, I'll get to the data, the all-important
data. So, you know, we went to

763
00:59:41,001 --> 00:59:48,001
archive for it. We used Twitter's 1%
Spritzer stream. And we got 52 months

764
00:59:48,001 --> 00:59:54,000
of data between October 2016 and December 2021.
46 million English language

765
00:59:54,000 --> 01:00:00,000
tweets. We have many more than that in
other languages. And 360,000 hashtags. So,

766
01:00:00,000 --> 01:00:04,000
we took all that data. We boiled it down
into a hashtag network, the co-occurring

767
01:00:04,000 --> 01:00:10,001
hashtags. And then we used this network
to calculate a lot of our measures. So,

768
01:00:10,001 --> 01:00:17,000
here's some of the you know, here's the
data community or shall we say the top

769
01:00:17,000 --> 01:00:23,001
10 most diverse hashtags within the
data community. There are about over 2

770
01:00:23,001 --> 01:00:27,001
,000 other hashtags in this community alone.
But you can see at the top, you have

771
01:00:27,001 --> 01:00:33,001
pretty familiar words there that you may
recognize. And then if we look across

772
01:00:33,001 --> 01:00:39,000
other communities of hashtags. So,
here's the data community, the beer community,

773
01:00:39,000 --> 01:00:43,000
the coffee community, and then the dog
community. At the top, you know, the most

774
01:00:43,000 --> 01:00:47,001
diverse, the most general hashtags are
very familiar terms. You would recognize

775
01:00:47,001 --> 01:00:51,001
most all of them. And then at the bottom,
the least diverse hashtags, the most

776
01:00:51,001 --> 01:00:56,000
specific, you know, I've never seen these
before. They look like they sort of

777
01:00:56,000 --> 01:01:02,001
make sense in the community.
But they're very narrowly used. And there's a

778
01:01:02,001 --> 01:01:09,000
beautiful tail here, which is based on
our findings. Hashtag love happens to be

779
01:01:09,000 --> 01:01:15,000
the most diverse hashtag, according to
our method, according to the archive.org

780
01:01:15,000 --> 01:01:19,000
data that we have. So, that's a beautiful
story there. And you can see some of

781
01:01:19,000 --> 01:01:24,001
the related communities to the love community.
These are the most

782
01:01:24,001 --> 01:01:30,000
tightly corresponding communities.
And, you know, not shown are thousands of

783
01:01:30,000 --> 01:01:34,000
other communities and hundreds of
thousands of other hashtags. But you can

784
01:01:34,000 --> 01:01:39,000
certainly, you know, dive into it.
So, what can be used? It can be used as a

785
01:01:39,000 --> 01:01:44,000
hashtag recommender. So, if hashtag Dalmatian,
maybe also hashtag dog. But the

786
01:01:44,000 --> 01:01:49,000
idea of this general to specific ordering
may I suggest that you don't want to do

787
01:01:49,000 --> 01:01:53,001
if hashtag dog, maybe also hashtag Dalmatian,
that doesn't really work. You can

788
01:01:53,001 --> 01:01:58,000
look at hierarchical hashtag topic models.
So, we know that hashtag AI is part of

789
01:01:58,000 --> 01:02:01,001
the data community. So, that's, you know,
good information to know. We also know

790
01:02:01,001 --> 01:02:06,000
that it is the data community. It can do
social science investigations. What does

791
01:02:06,000 --> 01:02:10,001
hashtag Minnesota mean?
And when was that meant? It certainly was more

792
01:02:10,001 --> 01:02:16,000
geographical prior to May of 2020 with
the murder of George Floyd in Minneapolis

793
01:02:16,000 --> 01:02:21,001
when Minnesota was now associated with
the Black Lives Matter movement. And you

794
01:02:21,001 --> 01:02:26,001
can use other platforms too. We showed Twitter.
But, you know, any tags will do

795
01:02:26,001 --> 01:02:31,000
for this. So, thank you for your time.
I appreciate it. And thank you to the

796
01:02:31,000 --> 01:02:35,001
presenters again. And if you're interested,
these are the names of the papers

797
01:02:35,001 --> 01:02:37,000
relevant. Thank you.

798
01:02:38,000 --> 01:02:43,001
Thank you so much, Spencer. We have some
questions that are coming in. And if you

799
01:02:43,001 --> 01:02:47,000
in the audience have additional questions,
please do drop them into the Q&A. The

800
01:02:47,000 --> 01:02:53,000
one thing that I was just struck by that data,
beer, coffee, dog hashtags are

801
01:02:53,000 --> 01:02:58,001
life. So, hold tight there for a second.
Before we go on, I want to acknowledge,

802
01:03:00,000 --> 01:03:04,001
I'm remiss in not acknowledging Art Rhino,
who is in the audience today, that

803
01:03:04,001 --> 01:03:09,001
fantastic video that we just watched.
Art is here watching along with us. So,

804
01:03:09,001 --> 01:03:15,001
thank you, Art, for that submission and
for your creative work. Up next in our

805
01:03:15,001 --> 01:03:20,001
show is another video. And another
presenter who is also in the audience. So,

806
01:03:21,000 --> 01:03:25,000
we're going to hear from Jim Salmons,
who is going to tell us about his Internet

807
01:03:25,000 --> 01:03:30,001
Archive-enabled journey as a digital
humanities citizen science. So, Caitlin,

808
01:03:30,001 --> 01:03:36,001
let's roll. Hi, I'm Jim Salmons.
And this lightning talk is a whirlwind trip

809
01:03:36,001 --> 01:03:41,000
through my post cancer journey of
rebirth as a digital humanities citizen

810
01:03:41,000 --> 01:03:46,000
scientist and how the inspiration of and
access to the Internet Archive made that

811
01:03:46,000 --> 01:03:52,001
possible. Starting in 2012 through 2014,
both my wife, Timon Babitzky, and I had

812
01:03:52,001 --> 01:03:57,001
terrifying cancer battles that we
fortunately survived. To pay it forward in

813
01:03:57,001 --> 01:04:03,000
celebration of our 25th wedding anniversary,
we funded the digitization of the 48

814
01:04:03,000 --> 01:04:07,001
issues of the Apple computer-focused
Soft Talk magazine into the Internet

815
01:04:07,001 --> 01:04:13,001
Archive. Soft Talk has a special place in
my heart as I was a reader, advertiser,

816
01:04:14,000 --> 01:04:19,000
writer, and during the time of its
explosive growth, an executive at Soft Talk

817
01:04:19,000 --> 01:04:23,000
Publishing where I designed and helped
develop the software that ran the back

818
01:04:23,000 --> 01:04:28,000
office production and advertising processes.
As part of our initial Soft Talk

819
01:04:28,000 --> 01:04:32,000
preservation project, Timon and I went
to the Midwest Scanning Center of the

820
01:04:32,000 --> 01:04:37,001
Archive where we saw firsthand and
participated in the amazing behind-the-scenes

821
01:04:37,001 --> 01:04:43,000
scanning service that puts digital
collections into the Internet Archive. Not

822
01:04:43,000 --> 01:04:47,000
wanting our contribution to end with
simply getting the digital edition of Soft

823
01:04:47,000 --> 01:04:51,000
Talk into the Archive, we followed up
with activity that was the true beginning

824
01:04:51,000 --> 01:04:55,000
of my rebirth as a digital
humanities citizen scientist.

825
01:04:55,001 --> 01:05:00,001
As I learned about the challenges of text
and data mining of digital collections

826
01:05:00,001 --> 01:05:05,000
within the cultural heritage domain,
I decided to create a ground truth storage

827
01:05:05,000 --> 01:05:10,001
format that would support an integrated
model of a magazine's complex document

828
01:05:10,001 --> 01:05:12,001
structures and content depiction.

829
01:05:13,001 --> 01:05:19,000
Between 2015 through 2019, I developed a
personal learning network of largely EU

830
01:05:19,000 --> 01:05:24,000
and UK-based mentors and collaborators to
support the development of the magazine

831
01:05:24,000 --> 01:05:30,001
GTS metadata format based on international
museum ontology standards. With

832
01:05:30,001 --> 01:05:36,000
posters and papers accepted to EU-based
digitization conferences and workshops,

833
01:05:36,001 --> 01:05:42,000
my work initially focused on the structure
and content of advertisements in Soft

834
01:05:42,000 --> 01:05:48,000
Talk. Everything came to crashing halt
when I suffered a devastating spinal cord

835
01:05:48,000 --> 01:05:54,000
injury in 2020. During the year and a half
of my rehab and recovery, the digital

836
01:05:54,000 --> 01:05:57,001
humanities domain saw
explosive growth in the use of

837
01:05:57,001 --> 01:05:59,000
machine learning technologies.

838
01:06:00,000 --> 01:06:05,000
As I reinvigorate my research,
I have expanded my focus from the Soft Talk

839
01:06:05,000 --> 01:06:09,001
collection to consider the challenges
of investigating the massive Internet

840
01:06:09,001 --> 01:06:14,001
Archive collection of computer magazines
consisting of tens of thousands of

841
01:06:14,001 --> 01:06:19,001
issues of publications in dozens of
languages published all around the world

842
01:06:19,001 --> 01:06:24,001
looking at the impact of computers and
emergence of the digital world we live in

843
01:06:24,001 --> 01:06:30,001
today. My initial exploration of this
expanded collection is focused on

844
01:06:30,001 --> 01:06:35,001
development of a ground truth dataset
of computer magazine talk or table of

845
01:06:35,001 --> 01:06:42,000
contents pages. Talks serve as a Suduco
puzzle-like set of hints about the

846
01:06:42,000 --> 01:06:47,000
document structures of a magazine.
This dataset will be invaluable for training

847
01:06:47,000 --> 01:06:52,001
machine learning models to help move
digitization pipelines from within page to

848
01:06:52,001 --> 01:06:58,000
whole document layout recognition.
My goal now is to stimulate collaboration

849
01:06:58,000 --> 01:07:03,001
between my EU and UK research friends with
new partners from Stanford Libraries,

850
01:07:04,000 --> 01:07:09,000
its AI lab, the Computer History Museum
and the Internet Archive to forge a

851
01:07:09,000 --> 01:07:13,001
research consortium to further preserve
and make accessible for scholarly

852
01:07:13,001 --> 01:07:18,000
research and public interest the computer
magazine's meta collection at the

853
01:07:18,000 --> 01:07:23,000
Internet Archive. On behalf of myself
and Tim Limobitsky, thank you to the

854
01:07:23,000 --> 01:07:28,001
archive and webinar organizers for inviting
me to present this lightning talk.

855
01:07:30,001 --> 01:07:36,000
And thanks to you Jim for sharing your
inspiring story, your video, your work,

856
01:07:36,001 --> 01:07:41,001
and for sponsoring the digitization of
the materials that are now available to

857
01:07:41,001 --> 01:07:48,000
all at the Internet Archive. That your
table of contents, TOC, or talk work is of

858
01:07:48,000 --> 01:07:53,000
high interest to us. I know also from
my previous work with the Biodiversity

859
01:07:53,000 --> 01:07:57,001
Heritage Library that those tables of
contents are really, that's where it's all

860
01:07:57,001 --> 01:08:01,000
at in terms of the structural
metadata for the for the article.

861
01:08:01,001 --> 01:08:03,001
So we'll be following up with learn a

862
01:08:03,001 --> 01:08:05,000
little bit more about what you're doing there.

863
01:08:06,000 --> 01:08:12,001
So up next I'd like to welcome Emmanuel
Tranos to the screen who's going to tell

864
01:08:12,001 --> 01:08:18,001
us about the relationship that exists
between the web and cities. This is a

865
01:08:18,001 --> 01:08:21,001
really interesting talk. I know you're
going to love it. So over to you Emmanuel.

866
01:08:24,000 --> 01:08:31,000
Chris, thank you so much for this.
Let me try my screen and also tell

867
01:08:31,000 --> 01:08:37,001
you how excited I am to be part of this
very cool session. So my name is Emmanuel

868
01:08:37,001 --> 01:08:42,000
Tranos. I'm a reader in quantitative human
geography at the University of Bristol

869
01:08:42,000 --> 01:08:47,001
and the Island Security Institute in the UK.
And I guess I'm all this rare breed

870
01:08:47,001 --> 01:08:52,001
of geographers who have a fascination
with the Internet. So today I'm going to

871
01:08:52,001 --> 01:08:59,000
give you a brief overview of our research
using data from the Internet Archive

872
01:08:59,000 --> 01:09:04,001
to understand the link between web and cities.
We're using such data to

873
01:09:04,001 --> 01:09:10,001
understand early Internet but also
other interesting geographs. So

874
01:09:10,001 --> 01:09:12,000
what data do we use?

875
01:09:13,001 --> 01:09:20,000
We're using data that is curated by the
British Library here in the UK. This is a

876
01:09:20,000 --> 01:09:26,000
dataset called the disk UK web domain dataset.
And this

877
01:09:26,000 --> 01:09:32,000
contains all the archived web
pages from the Internet Archive

878
01:09:32,000 --> 01:09:38,001
under the .
UK top level domain during the 1926 to 2012

879
01:09:38,001 --> 01:09:44,001
period. The British Library gives something
very clever here. They scanned the

880
01:09:44,001 --> 01:09:51,001
web text of all these archived web
pages and created a different, a

881
01:09:51,001 --> 01:09:58,000
further subset which only includes
these archived web pages which contain

882
01:09:58,001 --> 01:10:04,000
a UK postcode within their web text.
And I believe you can see the link to

883
01:10:04,000 --> 01:10:06,000
these datasets in the top.

884
01:10:06,001 --> 01:10:12,001
So we started our research with almost
half a billion of lines which look like

885
01:10:12,001 --> 01:10:19,001
this.
We know that this archived URL contains within

886
01:10:19,001 --> 01:10:26,001
its web text this UK postcode.
And this postcode refers to a very small area,

887
01:10:26,001 --> 01:10:32,000
almost a block in the UK.
And we also know the timestamp of this archival

888
01:10:32,000 --> 01:10:39,000
process. It happened on September,
the 9th of September of 2008. So what do we do

889
01:10:39,000 --> 01:10:46,000
with this data? Firstly,
we used this data to create a measure

890
01:10:46,000 --> 01:10:50,000
of the online content of local interest.

891
01:10:50,000 --> 01:10:57,000
And we must this data with a large
individual survey and were able

892
01:10:57,000 --> 01:11:03,001
to illustrate that the availability
of online content of local interest

893
01:11:03,001 --> 01:11:10,000
actually attracts individuals online.
We knew a lot about the fact that

894
01:11:10,000 --> 01:11:14,001
poor individuals, you know,
to connect to their internet to spend more time

895
01:11:14,001 --> 01:11:19,000
online. But this was the first time we
were able to say something about the

896
01:11:19,000 --> 01:11:24,000
actual factors that attract individuals
to spend more time online. And

897
01:11:24,000 --> 01:11:28,001
importantly, to do this at a very local scale.

898
01:11:29,001 --> 01:11:36,001
At a different scale, at a different study,
we used such data in order to

899
01:11:36,001 --> 01:11:43,000
understand economic clusters. And when I
say economic clusters, I refer to these

900
01:11:43,000 --> 01:11:49,000
neighborhoods within cities that host very,
you know, specific economic

901
01:11:49,000 --> 01:11:53,001
activities, these very specialized
neighborhoods on specific economic activities.

902
01:11:54,000 --> 01:12:01,000
We focused on Sordid. This is a very well
known tech cluster here in London. And

903
01:12:01,000 --> 01:12:07,001
we used archived web data in
order to map the evolution of

904
01:12:07,001 --> 01:12:14,001
this economic cluster over space and time.
But importantly, to map the evolution

905
01:12:14,001 --> 01:12:20,001
of this cluster, also in terms of the
types of economic activities that took

906
01:12:20,001 --> 01:12:26,000
place, you know, within this cluster.
And we're able, you know, to extract

907
01:12:26,000 --> 01:12:31,000
meaningful types of economic activities,
much more detailed than official data

908
01:12:31,000 --> 01:12:33,000
will have enabled us to do so.

909
01:12:35,000 --> 01:12:40,000
Again, we're changing the scale of analysis,
and we're moving, you know, from a

910
01:12:40,000 --> 01:12:47,000
small neighborhood in London to the
whole of the UK. We utilize, we

911
01:12:47,000 --> 01:12:53,001
employ such archived web data
in order to test the economic

912
01:12:53,001 --> 01:13:00,001
effects that the early adoption of web
technologies can generate to regions

913
01:13:00,001 --> 01:13:07,000
here in the UK. So we're able to build,
you know, measures of the volume

914
01:13:07,000 --> 01:13:14,000
of online commercial content back from 2000.
And

915
01:13:14,000 --> 01:13:20,001
we're able to link, you know,
these measures from 2000 to specific regions

916
01:13:20,001 --> 01:13:27,000
within the UK. We then utilize
econometric techniques, and we're able to

917
01:13:27,000 --> 01:13:34,000
illustrate an interesting lesson.
The volume of online content back from 2000,

918
01:13:34,000 --> 01:13:40,000
which to us represents the early adoption
of web technologies, is actually

919
01:13:40,000 --> 01:13:46,000
associated with positive economic
productivity effects. But these productivity

920
01:13:46,000 --> 01:13:51,001
effects are long-lasting. They are long-term,
you know, positive effects that

921
01:13:51,001 --> 01:13:58,001
regions who employ this technology early

922
01:13:58,001 --> 01:14:03,001
on are able to enjoy for longer,
for quite lengthy time periods.

923
01:14:05,000 --> 01:14:12,000
And last but not least, at the similar stage,
we used such archived web

924
01:14:12,000 --> 01:14:18,000
data, and more specifically,
HTML links between commercial websites

925
01:14:18,000 --> 01:14:25,000
to predict trade between different UK regions.

926
01:14:25,001 --> 01:14:31,000
More specifically, we're able to make
out of sample predictions using machine

927
01:14:31,000 --> 01:14:37,001
learning algorithms regarding this
regional trade clause. What is important?

928
01:14:38,000 --> 01:14:45,000
Because there is hardly any data for trade
between regions, such as small estate.

929
01:14:45,001 --> 01:14:49,001
So by using these three-level variable,
you know, data that the Internet Archive,

930
01:14:49,001 --> 01:14:55,000
you know, collects for all of us,
we're able to make predictions for certain

931
01:14:55,000 --> 01:15:01,000
important policy elements and have local
authorities utilize these data that we

932
01:15:01,000 --> 01:15:08,000
generated. So all in all, using these
three-level variable data, we're able to

933
01:15:08,000 --> 01:15:13,000
map the evolution, and the geography
of the engagement with the Internet,

934
01:15:13,001 --> 01:15:18,000
especially at its early status. And trust me,
there is hardly any other data, you

935
01:15:18,000 --> 01:15:21,000
know, that can go so back in
time and also so granular.

936
01:15:21,001 --> 01:15:28,000
By doing that, we're able to draw important
lessons regarding the deployment of

937
01:15:28,000 --> 01:15:34,000
other future technologies. And also,
we're able to understand economic activities

938
01:15:34,000 --> 01:15:40,000
at a very detailed level, both in terms
of space and time, but also in terms of

939
01:15:40,000 --> 01:15:46,001
context that take place within and
between cities. All of these research

940
01:15:46,001 --> 01:15:52,001
can be found in various papers who have
published in public agencies as well that

941
01:15:52,001 --> 01:15:58,000
you can find on my website. Again,
thank you so much for having you here today.

942
01:16:00,000 --> 01:16:06,000
Thank you, Emmanuel, for sharing your research.
We do have a question for you and

943
01:16:06,000 --> 01:16:12,000
for Spencer, but we want to wrap up today
with a final video from our session. So

944
01:16:12,000 --> 01:16:18,000
we're going to welcome Tom Galli back to
the screen virtually, and he's going to

945
01:16:18,000 --> 01:16:24,000
share his research and his work into the
forgotten novels of the 19th century.

946
01:16:28,000 --> 01:16:32,001
I started reading old books back in
the 1960s when I was still a child.

947
01:16:33,001 --> 01:16:39,001
So how did I decide what books to read?
Well, we had a lot of books at home, and

948
01:16:39,001 --> 01:16:45,000
I read some of those. There was a small
public library nearby, and I like to

949
01:16:45,000 --> 01:16:51,000
browse the shelves there too.
And when I got my allowance, I would go to the

950
01:16:51,000 --> 01:16:56,000
local bookstore, look through those shelves,
and maybe buy a couple of

951
01:16:56,000 --> 01:16:57,001
paperbacks that caught my eye.

952
01:16:58,000 --> 01:17:02,000
The books I happened to see on those
shelves would shape my reading

953
01:17:02,000 --> 01:17:03,001
choices in the years to come.

954
01:17:05,000 --> 01:17:09,001
When I started reading 19th century novels,
I naturally gravitated towards

955
01:17:09,001 --> 01:17:16,000
authors and books that I had seen on those
shelves. Charles Dickens, Jane Austen,

956
01:17:16,000 --> 01:17:22,000
Nathaniel Hawthorne, Crime and Punishment,
The Adventures of Huckleberry Finn. In

957
01:17:22,000 --> 01:17:24,000
other words, the classics.

958
01:17:24,000 --> 01:17:30,001
It was only later, after I had access
to a large university library,

959
01:17:31,000 --> 01:17:36,000
that I discovered the vast number of other
novels published in the 19th century.

960
01:17:37,000 --> 01:17:42,001
Their pages were yellow and brittle,
and most had never been reprinted. But after

961
01:17:42,001 --> 01:17:48,000
I graduated from college, I could no
longer access those books, and I was stuck

962
01:17:48,000 --> 01:17:49,001
with the classics again.

963
01:17:51,000 --> 01:17:57,001
So, fast forward to around the year 2010.
Libraries around the world were now

964
01:17:57,001 --> 01:18:02,000
scanning their old books,
and the Internet Archive was making those books

965
01:18:02,000 --> 01:18:04,000
available online for free.

966
01:18:05,000 --> 01:18:10,001
Now, anyone in the world could read
all of those old novels, including the

967
01:18:10,001 --> 01:18:15,000
thousands that publishers had not
reprinted and marketed as classics.

968
01:18:16,000 --> 01:18:22,000
So, in 2021, just for fun,
I compiled a list of 19th century

969
01:18:22,000 --> 01:18:24,000
novels at the Internet Archive.

970
01:18:24,001 --> 01:18:30,001
I chose only novels that nobody seemed
to be reading anymore, the once popular

971
01:18:30,001 --> 01:18:34,000
fiction that had been overlooked
by the classics industry.

972
01:18:35,000 --> 01:18:40,001
You can find that list on the Internet
Archive's blog under the title, Forgotten

973
01:18:40,001 --> 01:18:45,000
Novels of the 19th Century.
I've enjoyed dipping into those

974
01:18:45,000 --> 01:18:47,001
books. Maybe you will too.

975
01:18:52,000 --> 01:18:58,001
So, we'll share that link out to Tom's
blog for everyone in the email follow

976
01:18:58,001 --> 01:19:04,000
-up, and I'm sure you'll find some new
things that you haven't read for a while.

977
01:19:04,001 --> 01:19:09,000
Thanks Duncan for sharing that out.
I would like to bring Spencer and Emmanuel

978
01:19:09,000 --> 01:19:14,000
back to the screen for a couple of questions,
and the first one's to Spencer.

979
01:19:14,001 --> 01:19:18,001
It's, have there been changes in how
Twitter users use hashtags over time,

980
01:19:18,001 --> 01:19:23,001
especially more recently? That's a really
good question. We haven't looked into

981
01:19:23,001 --> 01:19:28,000
it in that respect yet, but that is on
the docket. So, you know, again, I'll

982
01:19:28,000 --> 01:19:34,000
reference the Minnesota example where
prior to May 2020, all geography, I mean,

983
01:19:34,000 --> 01:19:38,001
has to do with states, capitals, right?
No mention of civil rights or anything,

984
01:19:39,000 --> 01:19:43,000
and then after May 2020,
very much non-geographical. Well, incidentally

985
01:19:43,000 --> 01:19:47,000
geographical, having to do with Black
Lives Matter and the murder of George

986
01:19:47,000 --> 01:19:49,001
Floyd. So, that's an example,
but it's a great question

987
01:19:49,001 --> 01:19:51,001
and something we intend to dive into.

988
01:19:52,000 --> 01:19:59,000
Thank you for that, and a follow-up
question for Emmanuel. Is your research

989
01:19:59,000 --> 01:20:03,001
methodology extensible to other geographic
areas, or would you need to change

990
01:20:03,001 --> 01:20:09,001
your approach for studying areas outside
the UK? No, it is absolutely extendable

991
01:20:09,001 --> 01:20:13,000
and replicable outside of the UK.

992
01:20:13,001 --> 01:20:19,001
The main difference is having, you know,
available, you know, data sets or

993
01:20:19,001 --> 01:20:23,000
subsets, you know, from pain and archive,
and actually this is one of the values

994
01:20:23,000 --> 01:20:29,000
of this data because these are openly
available data sources, and by using these

995
01:20:29,000 --> 01:20:34,000
data, what actually outperforms
some of the official data sources.

996
01:20:36,000 --> 01:20:41,000
That's fascinating. I know that there are
some additional questions, but I'm also

997
01:20:41,000 --> 01:20:47,001
sensitive about our time. So, what I'm
going to ask everyone to do is we're going

998
01:20:47,001 --> 01:20:51,000
to have everyone to share their contact
information. So, if you have additional

999
01:20:51,000 --> 01:20:56,001
questions for any of our presenters today,
maybe we can follow up offline or

1000
01:20:56,001 --> 01:21:03,001
after the session. So, thank you Spencer
and Emmanuel for your conversations

1001
01:21:03,001 --> 01:21:09,000
today and for taking some time to field
some questions. Thank you much, and also

1002
01:21:09,000 --> 01:21:15,001
thanks to Jim and to Tom for the videos
that they offered up, Jim and Timlin as

1003
01:21:15,001 --> 01:21:22,000
well. So, here we are at the end.
We've made it. I want to give

1004
01:21:22,000 --> 01:21:27,000
an acknowledgement to the some people
who helped guide this series from its

1005
01:21:27,000 --> 01:21:32,001
start, from inception, and we pulled
together an advisory group for our series,

1006
01:21:32,001 --> 01:21:38,001
and that included some real heavy hitters
in the digital humanities space. So, a

1007
01:21:38,001 --> 01:21:43,001
big thank you to Dan Cohen from Northeastern,
to Makiiba Foster from Broward

1008
01:21:43,001 --> 01:21:49,001
County Library, Mike Furlow from HathiTrust,
and from Harriet Green at Washington

1009
01:21:49,001 --> 01:21:55,001
University in St. Louis. They've really
helped shape the talks that we saw here

1010
01:21:55,001 --> 01:22:00,000
today and help give guidance on how we
should frame a conversation, a long tail

1011
01:22:00,000 --> 01:22:05,000
conversation, a longitudinal chat,
if you will, among our digital humanities

1012
01:22:05,000 --> 01:22:12,000
scholars. So, thanks to the panelists. I'm

1013
01:22:12,000 --> 01:22:16,001
going to go a little off script. I don't
know if Brewster is still on the call, I

1014
01:22:16,001 --> 01:22:22,000
think that he is, but I would like to
offer Brewster a chance if he's still

1015
01:22:22,000 --> 01:22:27,000
available and if he's near his computer
to come on screen and give us a little

1016
01:22:27,000 --> 01:22:34,000
wrap up. What do you think of everything
that you've seen here today and across

1017
01:22:34,000 --> 01:22:41,000
the series? Oh, this is so inspiring
and so fun, and it's just to be able to

1018
01:22:41,000 --> 01:22:46,001
see that the light and humorous nature of this,
but as well as answering these

1019
01:22:46,001 --> 01:22:52,000
real questions. Of course,
I spring to the, we have this more to

1020
01:22:52,000 --> 01:22:58,000
share with you. We have periodicals that
have that electronic data processing

1021
01:22:58,000 --> 01:23:00,000
schools, institute schools.

1022
01:23:00,001 --> 01:23:06,001
Anyway, I'm so glad that this is going on.
I think the lightning

1023
01:23:06,001 --> 01:23:08,000
format at least works for me.

1024
01:23:09,000 --> 01:23:12,001
So, thank you very much for pulling
this together, Chris and everyone.

1025
01:23:13,001 --> 01:23:17,001
Yeah, thanks. I think this, the lightning
format, it's nice to have the variety,

1026
01:23:17,001 --> 01:23:22,001
right? You can do some deep dives on some
individual topics and then get a broad

1027
01:23:22,001 --> 01:23:26,001
overview of the wealth of materials and
the wealth of research that's happening

1028
01:23:26,001 --> 01:23:30,001
to in and around the collections
at the Internet Archive.

1029
01:23:31,000 --> 01:23:34,001
What I would say as we wind down
here today is this is the start of a

1030
01:23:34,001 --> 01:23:38,000
conversation. Let's think of this as the
start of a conversation, not as the end

1031
01:23:38,000 --> 01:23:42,000
of one with this series. Part of what we
wanted to do in bringing this together

1032
01:23:42,000 --> 01:23:46,001
was to one, raise awareness just that
there are digital humanities projects that

1033
01:23:46,001 --> 01:23:50,000
are actively using the collections and
the infrastructure at the Internet

1034
01:23:50,000 --> 01:23:53,001
Archive. And so what we were hoping was
to bring some of those people together,

1035
01:23:54,000 --> 01:23:58,001
provide some visibility into the work
that everyone is doing and start that

1036
01:23:58,001 --> 01:24:04,000
conversation so that we can do more together.
So, to wind down here today on

1037
01:24:04,000 --> 01:24:08,001
behalf of the Internet Archive and our
presenters today, I'd like to thank you

1038
01:24:08,001 --> 01:24:13,000
all for your time and your participation
and a big thanks to everyone who's

1039
01:24:13,000 --> 01:24:17,000
joined in the multiple sessions that
we've had along this series. Thanks

1040
01:24:17,000 --> 01:24:18,001
everyone. Have a great day.